canary: detect provider API schema drift; release v0.7.0

The export reads ChatGPT's and Claude's undocumented internal web APIs,
which can change shape without notice; the worst failure for a backup tool
is a silent one (skipped/mis-parsed content with no error). Add a `canary`
command + BaseProvider.check_drift() (overridden by ChatGPT/Claude) that
fetches one listing page + one conversation per provider and asserts only
the normalizer's load-bearing fields — not the full response shape, which
churns harmlessly. The top silent risk it guards is a renamed retrieval-tool
author bypassing the hidden-content collapse. ERROR findings exit non-zero;
WARN findings are surfaced but non-fatal so a backup run is never blocked.

Also retires the in-app watch mode and headless StartOS direction from the
roadmap (tool stays a local, manually-run CLI) and updates docs.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.env.example

@@ -36,6 +36,31 @@ EXPORT_DIR=./exports
 #   provider/year          → exports/claude/2024/file.md (ignores projects)
 OUTPUT_STRUCTURE=provider/project/year
 
+# What to do with content that was invisible in the provider's web UI
+# (file-retrieval tool dumps, hidden context like Custom Instructions).
+# These dumps can be 90% of a conversation's bytes. Options:
+#   placeholder  (default) → one-line placeholder with tool name and size
+#   full                   → keep everything (pre-v0.6.0 behavior)
+#   omit                   → drop entirely (still counted in the run summary)
+EXPORTER_HIDDEN_CONTENT=placeholder
+
+# Download conversation assets (images, audio) next to the Markdown, into a
+# media/ folder, and inline them. Options:
+#   images  (default) → images only
+#   all               → also audio/voice clips and other files
+#   off               → keep text placeholders, download nothing
+# Downloaded media is uploaded to Joplin as resources on the next `joplin` run.
+EXPORTER_DOWNLOAD_MEDIA=images
+
+# Cap how many conversations are downloaded per export run (per provider).
+# Runs are resumable — a capped run continues where it stopped next time.
+# Keeps big backfills from looking like scraper traffic. Unset = unlimited.
+#MAX_CONVERSATIONS_PER_RUN=25
+
+# Seconds between consecutive API requests (small random jitter is added).
+# Default 1.0; set 0 to disable pacing.
+#REQUEST_DELAY=1.0
+
 # --- Joplin ---
 # Automate importing exported conversations into Joplin as notes.
 # Requires Joplin desktop running with the Web Clipper service enabled.

CHANGELOG.md

@@ -3,19 +3,83 @@
 All notable changes to this project will be documented here.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
-## [0.2.0] - Unreleased
+## [0.7.0] - 2026-06-28
-### Added
-- Joplin import automation: `joplin` command syncs exported Markdown files to Joplin as notes
-- Notebooks created automatically per provider+project (`ChatGPT - My Project`, etc.)
-- Re-running is safe: notes are updated, not duplicated (Joplin note ID stored in manifest)
-- `JOPLIN_API_TOKEN`, `JOPLIN_API_URL`, `JOPLIN_REQUEST_TIMEOUT` config variables
-- Configurable request timeout with clear error messages and actionable hints on timeout
-- `--project` filter on `export` and `list` commands (case-insensitive substring or `none`)
-- ChatGPT Projects support via `CHATGPT_PROJECT_IDS` env var
 
-## [0.1.0] - Unreleased
 ### Added
-- Initial implementation: ChatGPT and Claude export via internal web APIs
+- `canary` command — probes the ChatGPT/Claude web APIs for schema drift against the fields the parser actually depends on (one listing page + one conversation per provider), not the full response shape. Flags the silent-failure risks for a backup tool: a renamed retrieval-tool author bypassing the hidden-content collapse, a new `content_type`, drifted message fields, or non-empty Claude `attachments`/`files` the normalizer ignores. ERROR findings exit non-zero; WARN findings are surfaced but non-fatal. See `FUTURE.md` §10.
-- Markdown and JSON exporters
+- `doctor` now reports a real ChatGPT token-health check ("ChatGPT token active") based on the `/api/auth/session` `error` field. ChatGPT session tokens are JWEs whose `exp` is encrypted and unreadable client-side; the previous decode path could never yield an expiry. The honest signal is `error == "RefreshAccessTokenError"` (verified live), which means the session token is dead even though `expires`/`accessToken` still echo stale values. See `FUTURE.md` §9.
-- Local cache/manifest for incremental sync
+
-- CLI with export, list, cache, doctor, and auth commands
+### Fixed
+- `ChatGPTProvider._fetch_access_token` now checks the `/api/auth/session` `error` field and fails fast with a clear "refresh your token" message. Previously it returned the stale `accessToken` present on a dead session, producing a confusing downstream 401 instead of an actionable auth error.
+
+### Removed
+- `auth --from-browser` and the `browser-cookie3` dependency (`src/browser_tokens.py`). Browser cookie auto-extraction is not viable: modern Chromium App-Bound Encryption (Chrome 127+/current Brave) keys cookies off a SYSTEM-level layer that can't be decrypted off disk without Administrator rights and AV-flagged SYSTEM impersonation, and fails on Brave specifically (symptom: `Unable to get key for cookie decryption`). Auth is manual (DevTools wizard) only. See `FUTURE.md` §2 for the full rationale.
+
+## [0.6.0] - 2026-06-12
+
+First tagged release. The project was developed through several internal
+milestones (0.1.0–0.5.0) that were never published; their changes are
+consolidated here.
+
+### Added
+
+**Core export & sync**
+- ChatGPT and Claude export via their internal web APIs, with a local cache/manifest for incremental, resumable sync (only new or updated conversations are fetched; each is written to the manifest immediately).
+- Markdown and JSON exporters. Markdown rendering happens at exporter-write time (providers produce typed blocks; exporters render them), which keeps the door open for future Obsidian/HTML output.
+- CLI: `export`, `list`, `cache`, `doctor`, `auth`, `joplin`, `prune`.
+- `--project` filter on `export`/`list`/`joplin` (case-insensitive substring, or `none`).
+- ChatGPT Projects support via `CHATGPT_PROJECT_IDS` (project conversations are fetched separately from the default listing).
+
+**Joplin integration**
+- `joplin` command syncs exported Markdown to Joplin as notes; notebooks are created automatically and nested per provider+project. Re-running is safe — the Joplin note ID is stored in the manifest, so notes are updated, not duplicated.
+- `JOPLIN_API_TOKEN`, `JOPLIN_API_URL`, `JOPLIN_REQUEST_TIMEOUT` config, with actionable error messages on timeout/connection failure.
+
+**Rich content (typed blocks)**
+- Messages carry an ordered `blocks` list: text, code, thinking, tool_use, tool_result, citation, image_placeholder, file_placeholder, unknown.
+- ChatGPT voice mode: `audio_transcription` parts render as text; audio asset pointers render as `📎 File attached` placeholders with size/duration. Custom Instructions (`user_editable_context`/`model_editable_context`) now appear (were silently dropped) with a `> ℹ️ Hidden context` marker.
+- ChatGPT `execution_output`, `system_error`, and `tether_browsing_display` render as `tool_result` blocks (with `tool_name`, `is_error`, optional `summary`); transient browse spinners skip silently.
+- Defensive Claude block extraction (text/thinking/tool_use/tool_result/image, including nested-block flattening).
+- `_safe_fence` picks a backtick fence longer than any run in the content, so embedded triple-backticks can't corrupt rendering (verified live in Joplin).
+- Data-loss visibility: a `LossReport` summary at the end of every `export` run breaks down `unknown blocks` and `extraction failures` by raw type; `unknown` blocks render as visible `> ⚠️ Unsupported content` with the raw type and observed keys.
+
+**Invisible-content collapse (`EXPORTER_HIDDEN_CONTENT=full|placeholder|omit`, default `placeholder`; `export --hidden-content`)**
+- Messages that were invisible in the ChatGPT web UI are collapsed to one-line placeholders instead of exported in full. Two triggers: (a) tool-role retrieval dumps identified by `author.name` (`file_search`, `myfiles_browser`) — ChatGPT re-injects the full text of attached/project files on every tool run, measured at 86% of all content bytes across the three largest conversations and **not** hidden-flagged; (b) messages flagged `is_visually_hidden_from_conversation` (Custom Instructions). Code-execution results, web-search output, and all dialogue are untouched.
+- New `collapsed` block renders as `> 🔧 **Tool output** — file_search (15.0 KB) — omitted (EXPORTER_HIDDEN_CONTENT=full to keep)` (ℹ️ Hidden context variant for hidden-flagged messages). The `LossReport` gains a `collapsed by policy` section (count per origin + total KB) so the omission stays visible.
+
+**Claude Code session provider (`--provider claude-code`)**
+- Archives local agent transcripts from `~/.claude/projects/*/*.jsonl` (override with `CLAUDE_CODE_DIR`) — no tokens, no rate limits, no ToS exposure. Prose-only by default per the hidden-content policy: dialogue and deliverables kept; tool traffic grouped into one placeholder per activity run (`> 🔧 Tool output — 14 calls: Read ×9, Bash ×2 (86KB) — omitted`); thinking dropped (counted in the summary); subagent (`isSidechain`) and harness (`isMeta`, snapshots, command tags) records stripped. Titles from the last `ai-title` record; project from the working-directory basename. Joplin notebooks nest under the `AI-Claude` parent. Measured: 19MB of JSONL → 804KB of Markdown across 25 sessions.
+
+**Binary/image downloads (`EXPORTER_DOWNLOAD_MEDIA=images|all|off`, default `images`; `export --download-media`)**
+- Downloads ChatGPT conversation assets into a `media/` folder beside each export and inlines them — images become real `![](media/…)` embeds, files become local links. Two-hop fetch via `/backend-api/files/{id}/download` → signed URL (verified live). `all` also pulls voice-mode audio (≈162MB of clips in this archive, transcripts already in text — hence the images-only default). Idempotent (skips assets already on disk), atomic writes, `600` perms. Expired/missing assets (old generated images 404) keep their placeholder and are counted as `media failed` — never fatal.
+- Joplin resource upload: the `joplin` command uploads downloaded media as resources and rewrites `media/…` links to `:/resourceId`, so images render inside notes (verified live: resources created with correct mime/size and linked to the note). Resource IDs are tracked per-conversation in the manifest, so re-syncs reuse them instead of duplicating.
+
+**Token setup & throughput**
+- `auth --from-browser [brave|chrome|chromium|edge|firefox]`: extracts ChatGPT session-token cookies and the Claude `sessionKey` straight from a local browser's cookie store (via `browser-cookie3`; defaults to Brave), validates each against the live API, and writes `.env`. Tokens that fail validation never overwrite a working entry; clear fallback when the browser is absent or the cookie DB is locked.
+- Session limiter: `--max-conversations N` / `MAX_CONVERSATIONS_PER_RUN` cap downloads per run (per provider); capped-out conversations are reported as "deferred" and resumed next run.
+- Polite pacing: `REQUEST_DELAY` (default 1.0s, ±25% jitter, `0` disables) between consecutive API requests, with the existing 429 backoff as the reactive net.
+
+**Re-rendering**
+- `export --force` re-exports every conversation even if cached and unchanged, so the whole archive can be re-rendered after a formatting/feature change without `cache --clear`. Runs as a tracked campaign (a stamped start time in the manifest): each run re-renders the least-recently-exported conversations, the "still to go" count shrinks toward zero, finished providers do no further work, and the campaign auto-completes (reporting "Force re-render complete"). Combines with `--max-conversations` to spread the work across runs.
+- `mark_exported` now preserves `joplin_note_id` / `joplin_synced_at` / `joplin_resources` across re-exports (refreshing `exported_at`), so a re-export followed by `joplin` updates the existing notes instead of creating duplicates.
+- Fix: `.env` is now loaded at the very start of every command, so `CACHE_DIR` and `LOG_FILE` are honored (previously the cache silently used `./cache` regardless of `CACHE_DIR`).
+
+**Archive hygiene**
+- `prune` deletes export files not referenced by the manifest (old-layout trees, no-ID orphans), with listing, confirmation, `--dry-run`, `-y`, and empty-directory sweep. Refuses to run when the manifest references no files (so `cache --clear` + `prune` can't wipe the archive). First live run removed 420 stale files (9.4 MB).
+- `doctor` verifies manifest ↔ disk integrity: every recorded `file_path` must exist on disk.
+
+### Changed
+- ChatGPT role filter that dropped `tool`/`system` messages is **lifted**; all roles route through normal extraction (truly empty messages skip via the empty-content guard).
+- `BaseProvider.normalize_conversation` accepts an optional `LossReport` parameter.
+- `ChatGPTProvider.normalize_conversation` reads `conversation_id` as a fallback for `id` (live detail responses use `conversation_id`; fixtures use `id`).
+- `Config` validates `EXPORTER_HIDDEN_CONTENT`, `EXPORTER_DOWNLOAD_MEDIA`, `MAX_CONVERSATIONS_PER_RUN`, and `REQUEST_DELAY`, and logs the active values at startup.
+- New dependency: `browser-cookie3==0.20.1`.
+
+### Fixed
+- Custom Instructions (`user_editable_context`/`model_editable_context`) were silently dropped from every conversation (parts-vs-direct-fields mismatch).
+
+### Migration
+- JSON exports: messages contain typed `blocks` and may omit the legacy `content` field — external consumers should prefer `blocks`.
+- To re-render existing exports with the current rendering and collapse policy: `python -m src.main cache --clear` then `python -m src.main export`, followed by `joplin` to update notes (and upload media resources). Per-conversation message counts may increase as previously-dropped Custom Instructions, image-only turns, and tool-only turns now appear.
+
+### Test suite
+- 264 tests, all passing.

FUTURE.md

@@ -7,21 +7,455 @@ of these additions straightforward.
 **Completed:**
 - v0.1.0 — Core export: ChatGPT + Claude, incremental sync, Markdown + JSON output
 - v0.2.0 — Joplin import automation (`joplin` command, create/update notes, notebook auto-creation)
+- v0.4.0 — Rich content support: typed message blocks (text, code, thinking, tool_use, tool_result, image_placeholder, file_placeholder, unknown); ChatGPT voice transcripts as text + audio placeholders; Custom Instructions extraction; data-loss visibility via `LossReport` summary and visible `unknown` blocks
+- v0.5.0 — Nested Joplin notebooks, date-prefixed note titles, flat year folders
+- v0.6.0 — Collapse tool retrieval dumps & hidden context (`EXPORTER_HIDDEN_CONTENT` policy; roadmap item 1); session limiter + request pacing (`MAX_CONVERSATIONS_PER_RUN`, `REQUEST_DELAY`; roadmap item 6)
 
 ---
 
-## Export `--force` Flag (v0.2.x)
+# Roadmap (decided 2026-06-12)
 
-Add `--force` to the `export` command to re-export already-cached conversations
+Priorities reflect the tool's primary purpose — a trustworthy backup so that
-without permanently clearing the entire manifest. Useful for re-generating files
+conversation data is not lost if a provider account is ever closed — plus the
-after changing the Markdown template or output structure.
+day-to-day friction of the weekly ChatGPT token refresh. The tool stays a
+local, manually-run CLI; the headless/StartOS direction was dropped
+2026-06-28 (see #7 and #8), which also retires the token-freshness problem
+(manual refresh is sufficient).
 
-Implementation: pass a `force=True` flag to `cache.get_new_or_updated()`, which
+**Now (in order):**
-returns all conversations regardless of cache state when force is True.
+1. ~~Collapse tool retrieval dumps & hidden context~~ — **shipped in v0.6.0**
+   (full-archive `export --force` re-export completed 2026-06-13)
+2. ~~Brave cookie auto-extraction~~ — **shipped in v0.6.0, removed afterward.**
+   Not viable: modern Chromium App-Bound Encryption (Chrome 127+/current Brave)
+   needs admin + SYSTEM impersonation that AV flags as credential theft, and
+   fails on Brave specifically. Auth is manual (DevTools) — see entry below.
+3. ~~Claude Code session provider~~ — **shipped in v0.6.0**
+4. ~~Archive hygiene — `prune` + `doctor` integrity check~~ — **shipped in v0.6.0**
 
-Current workaround: `python -m src.main cache --clear` then re-run export.
+**Soon, but later:**
 
-## Joplin `--force` Flag (v0.2.x)
+5. ~~Binary content downloads~~ — **shipped in v0.6.0**
+6. ~~Per-session download limiter + polite pacing~~ — **shipped in v0.6.0**
+7. ~~Scheduled / watch mode~~ — **dropped 2026-06-28**; the tool stays a
+   manually-run CLI, so no in-app polling loop is needed
+8. ~~StartOS service packaging~~ — **dropped 2026-06-28**; the local CLI is
+   sufficient (source convos live in the cloud and can be re-downloaded;
+   Joplin already syncs encrypted to an offsite S3 provider). Dropping this
+   also retires the headless token-freshness problem — manual weekly refresh
+   is fine.
+
+**Active:**
+9. ~~Surface remaining token validity on `doctor`~~ — **IMPLEMENTED 2026-06-28**
+   via the `/api/auth/session` `error` field (not `expires`). See §9.
+10. ~~Provider API-drift detection~~ — **IMPLEMENTED 2026-06-28** as the
+    `canary` command. See §10.
+
+**Deprioritized** (entries kept at the bottom of this file; revisit on
+demand): `--force` flags, per-conversation cache reset, official export-ZIP
+fallback, o1/o3 reasoning reclassification, Obsidian output, search command.
+Additional web providers (Gemini/Grok/Perplexity) are explicitly out of
+scope — no significant usage to archive.
+
+---
+
+## 1. Collapse Tool Retrieval Dumps & Hidden Context — SHIPPED v0.6.0
+
+**Implemented 2026-06-12** as designed below, with one scoping correction
+from live recon: the retrieval dumps are NOT flagged
+`is_visually_hidden_from_conversation` — `author.name == "file_search"` is
+the discriminator (the hidden flag only marks Custom Instructions and small
+system stubs). Verified live: worst files shrink 93% (524KB → 36KB).
+Full-archive re-export (the `export --force` campaign) + Joplin re-sync
+completed 2026-06-13.
+
+**Problem (measured 2026-06-12 against a full fresh export):** 45% of the
+entire 11.2MB archive (260 files) is tool-role messages; 29 files are
+majority tool-dump. Worst case: a 524KB conversation where 495KB (94%, 64
+messages) is ChatGPT's file-retrieval tool re-injecting the full text of
+the user's own attached documents, the same files dumped dozens of times
+per conversation. These messages were invisible in the ChatGPT web UI;
+they appear in exports because v0.4.0 lifted the role filter to fix silent
+data loss. Custom Instructions hidden-context blocks are a minor secondary
+case (~2KB, once per conversation) — the originally planned
+`EXPORTER_INCLUDE_HIDDEN_CONTEXT` toggle alone would not help: the worst
+file contains zero hidden-context blocks.
+
+**Fix: collapse, don't drop** (consistent with the no-silent-drop rule):
+
+- `EXPORTER_HIDDEN_CONTENT=full|placeholder|omit` env var, default
+  `placeholder`, plus a `--hidden-content` CLI override on `export`.
+- `placeholder` renders affected messages as one line with type and size:
+  `> 🔧 Tool output (file_search, 24KB) — omitted
+  (EXPORTER_HIDDEN_CONTENT=full to keep)`. Expected effect: archive
+  roughly halves; worst files shrink ~90%.
+- **Scope — collapse:** (a) tool-role retrieval dumps, identified by raw
+  `author.name` (`file_search`, `myfiles_browser`, …) in the API response
+  (the rendered Markdown only shows a generic "🔧 Tool" label, so the
+  decision must happen in the provider, not the renderer); (b) messages
+  flagged `is_visually_hidden_from_conversation`, including
+  `user_editable_context` / `model_editable_context` (Custom
+  Instructions) — subsumes the old suppress-hidden-context idea.
+- **Scope — keep at full size:** code-execution `tool_result` blocks and
+  web-search results; those are usually content the user wants.
+- Count collapsed messages in the post-export summary so the omission
+  stays visible (mirror the LossReport presentation, but as intentional
+  policy, not loss).
+
+Re-export workflow after shipping: `cache --clear` + `export` (same as
+the v0.4.0 migration).
+
+## 2. Brave Cookie Auto-Extraction — REMOVED (not viable)
+
+**Shipped v0.6.0 (2026-06-12), removed 2026-06-27.** `auth --from-browser`
+plus `src/browser_tokens.py` and the `browser-cookie3` dependency are gone.
+Auth is manual (DevTools) only. Do not re-attempt without a fundamentally
+different mechanism (see below).
+
+**Why it doesn't work.** Modern Chromium browsers encrypt cookies on Windows
+with **App-Bound Encryption** (Chrome 127+, July 2024; current Brave).
+Cookies are written with a `v20` prefix and keyed off a secret wrapped in a
+**SYSTEM-level** DPAPI layer plus app validation. `browser-cookie3` only
+knows the legacy `v10`/DPAPI key, so its AES-GCM MAC check fails — the exact
+symptom hit in the field:
+
+```
+ChatGPT: Could not read brave cookies for chatgpt.com: Unable to get key for cookie decryption.
+```
+
+Decrypting `v20` at all requires unwrapping the SYSTEM layer, which means
+running as SYSTEM (e.g. a PsExec-style service) — i.e. **Administrator
+rights** and behavior that AV/EDR flags as infostealer activity. The one
+maintained Python option (`rookiepy`) needs admin from Chrome v130+, was
+**archived 2026-06-07**, and has an unresolved bug where **Brave returns 0
+cookies** even after the key is retrieved. ABE is *designed* to stop exactly
+this, so no off-disk reader is a reliable, non-invasive fit.
+
+**If ever revisited:** the only non-admin path is Chrome Remote Debugging
+(launch the browser with `--remote-debugging-port`, read cookies via
+`Network.getAllCookies` — the running browser decrypts for you). Heavier and
+intrusive; not worth it for a weekly token refresh that takes 30 seconds by
+hand. With the headless/StartOS direction dropped (#8), manual DevTools
+refresh is the accepted approach — no automated extraction is needed.
+
+## 3. Claude Code Session Provider — SHIPPED v0.6.0
+
+**Implemented 2026-06-12** as designed below (`src/providers/claude_code.py`,
+`--provider claude-code`, `CLAUDE_CODE_DIR` override). Additional findings
+during implementation: `isSidechain` records are subagent transcripts (skipped),
+`isMeta` marks harness-generated user records (skipped), and listing/normalized
+`updated_at` must both use file mtime or the cache would re-export every
+session every run. First export: 25 sessions, 19MB JSONL → 804KB Markdown.
+
+Archive local Claude Code session transcripts. No tokens, no rate limits,
+no ToS risk — the data is already on disk but lives in a single JSONL per
+session that Claude Code may clean up, and it contains deliverables
+(reviews, plans, analyses) that exist nowhere else.
+
+Decisions (2026-06-12):
+- **Rendering: prose-only.** Keep user prompts and assistant text
+  (including full deliverable write-ups); collapse tool activity to
+  one-line placeholders (`> 🔧 Tool activity — 14 calls (Read ×9, Bash ×2),
+  86KB — omitted`); **exclude thinking blocks**.
+- **Joplin: sync enabled.** Each coding project becomes a notebook nested
+  under an **"AI-Claude"** parent notebook (nested-notebook support shipped
+  in v0.5.0).
+
+Data facts (measured 2026-06-12):
+- Source: `~/.claude/projects/<munged-cwd>/<session-uuid>.jsonl`.
+  Currently 29 sessions, 18.8MB total, largest 3.8MB.
+- Representative 2.1MB session: tool_result 436KB, tool_use 122KB,
+  thinking 104KB, dialogue prose only ~24KB (~4%) — collapsing tool
+  activity is what makes these exports readable.
+- Record types: `user` / `assistant` (Anthropic-style `message.content`
+  block arrays) plus harness records: `ai-title` (use for note title and
+  filename slug), `last-prompt`, `file-history-snapshot`, `attachment`,
+  `permission-mode`, `system` (skip). Strip harness noise from user
+  messages (`<local-command-caveat>`, `<command-name>` blocks).
+
+Implementation shape: new `src/providers/claude_code.py` implementing the
+`BaseProvider` interface — `list_conversations` scans project dirs,
+`get_conversation` parses the JSONL, `normalize_conversation` maps onto the
+existing block schema (content is already block-shaped: text / tool_use /
+tool_result / thinking). Incremental sync via file mtime/size recorded in
+the existing manifest. Project name derives from the munged cwd dirname.
+
+## 4. Archive Hygiene: `prune` Command + Manifest Integrity — SHIPPED v0.6.0
+
+**Implemented 2026-06-12** as designed below, plus an empty-manifest guard
+(refuses to prune right after `cache --clear`). First live run removed 420
+stale files (9.4 MB, old-layout trees + `_.md` orphans); doctor now reports
+manifest↔disk integrity (293/293 after the run).
+
+A backup is only trustworthy if the on-disk tree matches the manifest.
+Observed 2026-06-12: pre-v0.5.0 layout trees (`tspc-expertcouncil/2025/`)
+and `_.md` no-ID orphans (from the empty-conversation-id bug fixed in
+v0.4.1) sit alongside current exports and would double-sync into Joplin.
+
+- `prune` command: delete export files not referenced by the manifest.
+  `--dry-run` (default off, but always print the list before deleting)
+  shows what would be removed and why (old layout / orphan / unknown).
+- `doctor` extension: verify every manifest entry's `file_path` exists on
+  disk; report missing files (re-export candidates) and unreferenced files
+  (prune candidates).
+
+## 5. Binary Content Downloads — SHIPPED v0.6.0
+
+**Implemented 2026-06-12** (`src/media.py`, `EXPORTER_DOWNLOAD_MEDIA`,
+`download_asset`/`parse_asset_file_id` on the ChatGPT provider, Joplin
+`create_resource` + `upload_media_and_rewrite`). Live recon settled the
+download mechanism: `GET /backend-api/files/{id}/download` returns a signed
+`download_url`; a second GET yields the bytes (works for user uploads;
+older AI-generated images 404 — expired server-side, handled gracefully).
+Asset refs come in three shapes — `sediment://file_…`,
+`sediment://<hash>#file_…#p_N.png` (generated), `file-service://…`. Archive
+scan: 14 images (4 uploads / 10 generated) + 556 audio clips ≈162MB, so
+images-only is the default and audio is opt-in via `all`.
+
+Original notes below.
+
+**Priority note (2026-06-12): "later, but soon" — under the
+backup-if-account-closes goal, embedded images are part of the data that
+would be lost; placeholders alone don't preserve them.**
+
+v0.4.0 ships placeholders for images and audio assets but does not download
+the binary content. The `_safe_fence`-wrapped placeholders include the asset
+reference (`sediment://...` or `file-service://...`), MIME type, size, and
+duration where available; the actual bytes are not preserved.
+
+Next steps:
+- Download attached images alongside the Markdown export, save under a
+  `media/` sibling directory with a stable filename derived from the asset
+  reference.
+- Replace `image_placeholder` rendering with an inline `![](relative/path)`
+  reference once the file is on disk.
+- Joplin integration: upload binaries as Joplin resources via `POST /resources`,
+  rewrite the rendered Markdown to use `:/resourceId` references, and track
+  the resource ID in the cache manifest so re-syncs stay idempotent.
+- DALL-E images on the assistant side: not observed in this user's data; the
+  code path exists (`source = "model_generated"`) but is untested.
+
+The block-level schema is already in place — only the file-fetch + rewrite
+layer needs to be added. See the `image_placeholder` and `file_placeholder`
+block definitions in `src/blocks.py`.
+
+## 6. Per-Session Download Limiter — SHIPPED v0.6.0
+
+**Implemented 2026-06-12** as designed below: `--max-conversations N` /
+`MAX_CONVERSATIONS_PER_RUN` session cap with deferred-count reporting, and
+`REQUEST_DELAY` pacing (default 1.0s ±25% jitter) in `BaseProvider._request`.
+Verified live: a 2-pending run with cap 1 exported one, deferred one, and
+the re-run picked it up.
+
+Cap how many conversations are downloaded in a single `export` run so the
+tool never hammers the ChatGPT/Claude internal APIs with a large burst —
+most importantly on the very first export, which otherwise fetches the
+entire conversation history in one session. Because every run is resumable
+(the manifest records each conversation immediately), a capped run simply
+exports the first N pending conversations and the next run picks up where
+it left off. This keeps traffic looking like a human-paced session rather
+than a scraper, reducing the risk of rate limiting or account flags.
+
+Two complementary pieces:
+
+1. **Session cap** — `--max-conversations N` flag (and
+   `MAX_CONVERSATIONS_PER_RUN` env default). Implementation: in the
+   `export` command, slice the pending list after the cache filter:
+   `to_export = to_export[:n]`. On exit, print exported-vs-remaining
+   counts (reuse the message format from the 429 early-exit path) and
+   remind the user to re-run to continue.
+2. **Polite pacing** — `REQUEST_DELAY` env var (seconds, with small
+   random jitter) slept between per-conversation detail fetches in
+   `BaseProvider`, so even a capped run doesn't fire requests
+   back-to-back. The existing 429 backoff in `_request` stays as the
+   reactive safety net.
+
+Note: the conversation *listing* (paginated, 100/page) still runs in full
+each time so the cache comparison works — the cap applies to the heavy
+per-conversation detail fetches, which dominate request volume.
+
+This is a stepping stone to the StartOS service: a capped, politely-paced
+export — scheduled by the host (cron/StartOS), not an in-app loop — is the
+traffic profile a headless deployment needs.
+
+## 7. Scheduled / Watch Mode — DROPPED (2026-06-28)
+
+An in-app `watch`/scheduler loop is not worth building. Scheduling belongs to
+whatever hosts the tool: a user cron line locally, and on the long-term
+StartOS target the platform's own scheduling. Either way the tool only needs
+to do one capped, politely-paced `export` + `joplin` run and exit — which it
+already does. If cron ergonomics ever feel clunky, a thin `sync` subcommand
+that chains `export` then `joplin` for a single cron line is a trivial
+add-on, but the polling loop itself is off the roadmap.
+
+## 8. StartOS Service Packaging — DROPPED (2026-06-28)
+
+Not pursuing a headless StartOS service. The local, manually-run CLI is
+sufficient: the source conversations live in the providers' clouds and can be
+re-downloaded, and Joplin already syncs (encrypted) to an offsite S3 provider,
+so durability is covered without a server in the loop.
+
+Dropping this also retires the one genuinely hard sub-problem it carried —
+session-token freshness without a browser. There is no headless context to
+keep fresh; the weekly manual DevTools refresh is acceptable. (Local cookie
+extraction remains a dead end regardless — see #2.)
+
+## 9. Token Validity on `doctor` — IMPLEMENTED (2026-06-28)
+
+Shipped: `doctor` now adds a "ChatGPT token active" check via
+`ChatGPTProvider.session_health()` (reads `/api/auth/session`, passes iff
+`error` is falsy and `accessToken` is present), the never-working JWE/`exp`
+decode path was removed, and `_fetch_access_token` now fails fast on a set
+`error` instead of returning a stale token. Tests in
+`tests/test_providers.py::TestChatGPTSessionHealth`. Investigation trail
+below for the record.
+
+Goal: if it's cheap to tell how much longer a token will work, show it on
+`doctor`. Findings from live recon:
+
+- **Not readable from the token itself.** ChatGPT's `CHATGPT_SESSION_TOKEN`
+  is a **JWE** (header `{"alg":"dir","enc":"A256GCM"}`, `eyJ…` prefix is just
+  the encrypted protected header) — the `exp` claim is AES-256-GCM encrypted
+  with an OpenAI-only key, so it cannot be decoded client-side. Claude's
+  `sk-…` key is fully opaque. The existing `doctor` JWT-decode path therefore
+  never yields an expiry for the real tokens (falls to the "not decodable"
+  branch).
+- **`/api/auth/session` exposes an `expires`** (the provider already calls
+  this endpoint in `_fetch_access_token`; the response includes `expires`
+  alongside `accessToken`). Live value observed 2026-06-28:
+  `2026-09-26` — **~90 days out**. This contradicts both the code's ~7-day
+  assumption and the lived weekly-refresh cadence, so it is almost certainly
+  the NextAuth **rolling session window** (re-extended on every call), not
+  the point at which the pasted token actually 401s. Displaying it verbatim
+  would give false confidence.
+- **RESOLVED 2026-06-28 by a live 401 data point.** When the ChatGPT token
+  was actually dead (conversations API → 401), `/api/auth/session` still
+  returned **HTTP 200** with `expires: 2026-09-26` (~90 days out) — and that
+  `expires` *advanced* between two calls seconds apart (`04:20:08` → `04:30:37`).
+  So `expires` is a **rolling session window that rolls forward on every call
+  even for a dead token**; displaying it would actively lie. The same response
+  carried `error: "RefreshAccessTokenError"` and a stale `accessToken`.
+- **The real signal is `error`, not `expires` or `accessToken`.** On a healthy
+  token `error` is absent/null; when the session token is dead NextAuth can't
+  refresh and sets `error: "RefreshAccessTokenError"` while still echoing a
+  rolling `expires` and a stale `accessToken`. This is an exact, free, binary
+  health check.
+- **Design (ready to build):**
+  1. `doctor` ChatGPT check → read `/api/auth/session`; pass iff `error` is
+     falsy and `accessToken` present; on `RefreshAccessTokenError` report
+     "token expired — refresh". Drop the JWE/`exp` decode path (it can never
+     work) and do NOT surface `expires`.
+  2. Latent bug to fix alongside: `_fetch_access_token` reads `accessToken`
+     without checking `error`, so it proceeds with a stale token and yields a
+     confusing downstream 401 instead of a clear "refresh your token" message.
+     Check `error` there and fail fast.
+  3. Claude stays a 401-only signal (opaque `sk-`, no equivalent endpoint).
+
+## 10. Provider API-Drift Detection — IMPLEMENTED (2026-06-28)
+
+Shipped: the `canary` command + `BaseProvider.check_drift()` (overridden by
+ChatGPT and Claude). It fetches one listing page + one conversation per
+provider and asserts only the normalizer's load-bearing fields, emitting
+`DRIFT_OK/WARN/ERROR` findings (`src/providers/base.py`). Severity badges
+print as a Rich table; ERROR exits non-zero, WARN is non-fatal so a backup
+run is never blocked. Drift vocabularies (`_KNOWN_TOOL_AUTHORS`,
+`_HANDLED_CONTENT_TYPES`) live in `chatgpt.py` next to the collapse set they
+guard. Tests: `TestChatGPTDriftCanary`, `TestClaudeDriftCanary`,
+`TestCanaryCommand`. Verified live 2026-06-28 — both providers OK. Recon
+trail below for the record.
+
+## 10b. Provider API-Drift Detection — investigation (2026-06-28)
+
+The export depends on undocumented internal web APIs (ChatGPT/Claude) that can
+change shape without notice. The worst failure for a backup tool is *silent*:
+a response-schema change that makes the exporter skip or mis-parse content
+without erroring. `doctor` currently checks token validity, reachability, and
+manifest↔disk integrity — but not "does the provider's response still look
+like what the parser expects."
+
+To investigate: a lightweight schema/shape assertion on a known-good sample
+of each provider's listing + conversation-detail responses (presence and type
+of the fields the normalizers rely on), surfaced as a `doctor` check or a
+dedicated canary.
+
+**Live recon — Claude captured 2026-06-28 (ChatGPT pending a token refresh):**
+
+- **Dependency surface (assert ONLY these — see below for why):**
+  - listing item: `uuid`, `name`, `updated_at`/`created_at`, `project.name`.
+  - conversation detail: `uuid`/`id`, `name`, `created_at`, `updated_at`,
+    `project.name`, `chat_messages[]`.
+  - message: `sender` (`human`/`assistant`), `text` (string) or `content`
+    (list of typed blocks), `created_at`.
+- **Key finding — full-shape diffing is the wrong design.** Claude's
+  `settings` object is full of volatile internal codenames that churn
+  constantly: `enabled_bananagrams`, `enabled_sourdough`, `enabled_foccacia`,
+  `enabled_saffron`, `enabled_turmeric`, `enabled_monkeys_in_a_barrel`,
+  `paprika_mode`, `enabled_megaminds`, … A "any new/removed key = drift"
+  canary would fire on every UI experiment. The canary MUST target the
+  normalizer's load-bearing fields only, not the whole response. (Aligns with
+  the drop-noise-don't-retain-it principle.)
+- **Real Claude messages are flat `text`/`sender`** — in this archive every
+  message had a string `text` and NO `content` block list (0 rich blocks
+  observed). So `_extract_claude_blocks` / `_dispatch_claude_block` (tool_use,
+  thinking, image, …) is an **unexercised theoretical path**; drift there
+  can't be "caught" by a canary because it never runs on real data — it's a
+  safety net for if Claude ever switches to block content. The canary should
+  assert the flat shape and *warn if `content` ever appears as a list* (that
+  itself is the drift event that would activate the dormant code).
+- **Possible silent-loss spot (separate from drift):** Claude messages carry
+  `attachments` and `files` arrays (empty in this sample) that the normalizer
+  ignores entirely. If a user ever attaches files in Claude, they'd be
+  dropped without a LossReport entry. Worth a follow-up check.
+**Live recon — ChatGPT captured 2026-06-28:**
+
+- **Dependency surface (assert ONLY these):**
+  - listing item: `id`, `title`, `update_time`/`create_time`.
+  - conversation detail: `conversation_id`/`id`, `title`, `create_time`,
+    `update_time`, `mapping` (non-empty).
+  - mapping node: `message`, `children` (the tree walk depends on both);
+    message: `author.role`, `author.name`, `content.content_type`,
+    `content.parts`, `metadata.is_visually_hidden_from_conversation`.
+- **content_type vocabulary observed (all currently handled):** `text`,
+  `model_editable_context`, `multimodal_text`, `thoughts`, `code`,
+  `execution_output`, `reasoning_recap`, `user_editable_context`,
+  `tether_browsing_display`. A *new* content_type already degrades gracefully
+  (visible `unknown` block + WARNING + LossReport tally) — so content_type
+  drift is **already non-silent**. The canary just needs to confirm the known
+  set still parses to non-empty blocks.
+- **The genuinely silent drift risk — `author.name` collapse keys.**
+  `_COLLAPSE_TOOL_AUTHORS = {"file_search", "myfiles_browser"}`. Recon
+  confirms `file_search` is live (and `web.run`/`python` are correctly left
+  un-collapsed). If OpenAI renames `file_search`, the collapse **silently
+  stops** and the archive re-bloats with no error or LossReport entry. This is
+  the top canary target: assert that retrieval-dump tool authors are still
+  recognized, or at least flag unfamiliar `(role="tool", author.name)` pairs.
+- **Second silent risk — empty `content.parts`.** A `text` message whose
+  `parts` field is renamed/emptied yields zero blocks and is skipped with only
+  a debug log = silent loss. Canary should assert a sampled `text` message
+  produces a non-empty block.
+
+**Recon complete for both providers. Canary design (ready to build):** a
+`doctor` check (or dedicated `canary` command) that, per provider, fetches one
+listing page + one conversation and asserts the dependency-surface fields
+above by presence+type — NOT full shape (Claude `settings` codenames prove
+full-shape diffing is pure noise). Specific tripwires: (ChatGPT) unfamiliar
+`(tool, author.name)` pair and empty `parts` on a text message; (Claude)
+`content` appearing as a list, and non-empty `attachments`/`files`. Failures
+surface as a warning, never a hard error (a backup tool must still run).
+
+---
+
+# Deprioritized
+
+Kept for reference; not on the active roadmap.
+
+## Export `--force` Flag — SHIPPED v0.6.0
+
+Implemented 2026-06-12: `export --force` passes `force=True` to
+`cache.get_new_or_updated()`. Shipped alongside a `mark_exported` fix that
+preserves Joplin links across re-exports, so a forced re-render + `joplin`
+updates existing notes instead of duplicating them.
+
+## Joplin `--force` Flag
 
 Similarly, add `--force` to the `joplin` command to re-sync all cached
 conversations to Joplin regardless of whether they've been synced before.
@@ -30,7 +464,7 @@ Useful after making formatting changes to the Markdown exporter.
 Implementation: in `get_joplin_pending()`, return all entries that have a
 `file_path` when `force=True`, ignoring `joplin_synced_at`.
 
-## Per-Conversation Cache Reset (v0.2.x)
+## Per-Conversation Cache Reset
 
 Add `cache --reset --conversation <id>` to force re-export or re-sync of a
 single conversation without clearing the entire provider cache.
@@ -38,9 +472,7 @@ single conversation without clearing the entire provider cache.
 Current workaround: manually edit `~/.ai-chat-exporter/manifest.json` and
 delete the entry, then re-run export.
 
----
+## Official API Fallback
-
-## Official API Fallback (v0.3.0)
 
 If the unofficial internal web API approach breaks, migrate to official export
 file parsing as a fallback:
@@ -56,51 +488,17 @@ To add this: implement `src/providers/file_chatgpt.py` and
 `src/providers/file_claude.py`, then add `--input-file` flag to the
 export command to accept a pre-downloaded export ZIP or JSON.
 
----
+Deprioritized 2026-06-12: the official ChatGPT export does not cover what
+this user needs (project data), so it isn't a real fallback here.
 
-## Rich Content Support (v0.4.0)
+## Reclassify o1/o3 Reasoning Subparts
 
-Currently only text content is exported. Future versions should handle:
+v0.4.0 leaves dict parts inside `text` content_type messages with shape
+`{"summary": ..., "content": ...}` rendered as plain text (defensive — the
+shape was inferred from a code comment, not captured live). Once a real
+reasoning conversation is captured, reclassify these as `thinking` blocks.
 
-### Claude
+## Obsidian Vault Output
-- Artifacts (code, documents, HTML) — export as separate files, link from Markdown
-- Uploaded images — download and embed or link
-- Extended thinking/reasoning blocks — include as collapsible sections
-- Tool call results and web search citations — include as footnotes or appendices
-
-### ChatGPT
-- DALL-E generated images — download and embed or link
-- Code Interpreter outputs — export code and results
-- File attachments — download and reference
-- Voice transcripts — include as text
-
-Implementation note: the normalized message schema already includes a
-`content_type` field placeholder. When this work begins, extend the schema
-rather than replacing it. Non-text content already logs a WARNING when
-encountered so users can see what was skipped.
-
----
-
-## Scheduled / Watch Mode (v0.5.0)
-
-Add a `watch` command (or cron integration helper) to run exports automatically
-on a schedule:
-
-```bash
-python -m src.main watch --interval 6h   # poll every 6 hours
-```
-
-This would run `export` + `joplin` in sequence, then sleep. Alternatively,
-provide a `cron` command that prints the correct crontab line for the user's
-setup.
-
-Implementation: simple loop with `time.sleep()`, or emit a crontab entry
-string that calls the export and joplin commands in sequence. A `--once`
-flag would do a single run then exit (useful for cron itself).
-
----
-
-## Obsidian Vault Output (v0.5.0)
 
 Add an `obsidian` command (or `--target obsidian` flag) to sync exported
 conversations into an Obsidian vault directory. The current Markdown format
@@ -116,40 +514,17 @@ Obsidian. An `ObsidianSyncer` class (mirroring `JoplinClient`) would simply
 copy files to the vault directory and maintain a flat or nested folder
 structure matching the user's Obsidian setup. No API needed — just file I/O.
 
----
+## Token Expiry Notifications
 
-## Joplin Nested Notebooks (future)
+Moved to the active roadmap as §9 (Token Validity on `doctor`). The original
+"proactively notify before expiry" idea is blocked by the same finding: a
+reliable expiry time isn't available client-side (ChatGPT token is encrypted,
+Claude's is opaque, and `/api/auth/session`'s `expires` looks like a rolling
+window rather than the real refresh cadence). Any heads-up — a `doctor`
+line, an `expiry` subcommand, or a `notify-send` nudge — depends on first
+resolving the §9 open question of what signal is actually trustworthy.
 
-Currently notebooks are flat: `ChatGPT - My Project`. Joplin supports nested
+## Search Command
-notebooks via `parent_id`. A future option (`JOPLIN_NESTED_NOTEBOOKS=true`)
-could create a two-level hierarchy:
-
-```
-ChatGPT/
-  My Project/
-  No Project/
-Claude/
-  Budget Tracker/
-```
-
-Implementation: `get_or_create_notebook` would first find/create the provider
-notebook, then find/create the project notebook as a child.
-
----
-
-## Token Expiry Notifications (future)
-
-Proactively warn when a token is close to expiry (within 48h for ChatGPT),
-rather than only surfacing the warning at startup. Options:
-
-- Add an `expiry` subcommand that prints token status and exits non-zero if
-  any token is expired or expiring soon (useful in scripts/cron)
-- Send a desktop notification via `notify-send` (Linux) or `osascript` (macOS)
-  when a token is within 24h of expiry
-
----
-
-## Search Command (future)
 
 Add a `search` command to full-text search across all exported Markdown files:
 

README.md

@@ -79,7 +79,17 @@ Before anything else, validate your setup:
 ai-chat-exporter doctor
 ```
 
-This checks token presence, format, expiry, directory permissions, disk space, and live API connectivity. Fix any failures before proceeding.
+This checks token presence, format, token health (via the `/api/auth/session` `error` field), directory permissions, disk space, and live API connectivity. Fix any failures before proceeding.
+
+### Checking for API drift
+
+The exporter reads ChatGPT's and Claude's undocumented internal web APIs, which can change shape without notice. The worst failure for a backup tool is *silent* — a response change that makes the exporter skip or mis-parse content without erroring. Run the canary to catch that early:
+
+```bash
+ai-chat-exporter canary
+```
+
+It probes one conversation per provider and checks only the fields the parser depends on (a renamed retrieval-tool author that would bypass the content-collapse, a new content type, drifted message fields, ignored attachments). `OK` means the live schema still matches; `WARN`/`ERROR` tells you exactly what changed. Worth running before a large export or on a schedule.
 
 ---
 
@@ -87,12 +97,22 @@ This checks token presence, format, expiry, directory permissions, disk space, a
 
 Session tokens are how your browser stays logged in. This tool uses them to access your chat history on your behalf.
 
+Tokens are entered manually — copy them from your browser's DevTools and run the wizard:
+
+```bash
+ai-chat-exporter auth
+```
+
+The wizard detects your OS, shows the correct DevTools shortcut, and writes the values to `.env` without echoing them to the terminal.
+
+> **Why no auto-extraction from the browser?** Modern Chromium browsers (Chrome 127+, current Brave) encrypt cookies on Windows with **App-Bound Encryption**: the key is bound to the browser through a SYSTEM-level service, so reading cookies off disk requires Administrator rights and a PsExec-style SYSTEM impersonation that antivirus flags as credential theft — and it still fails on Brave specifically. There is no reliable, non-invasive way to pull these cookies automatically, so the manual DevTools flow below is the supported path.
+
 ### Token Lifetimes
 
 | Provider | Cookie Name | Lifetime | Expiry Detection |
 |----------|-------------|----------|-----------------|
-| ChatGPT | `__Secure-next-auth.session-token` | ~7 days | JWT `exp` claim (decoded automatically) |
+| ChatGPT | `__Secure-next-auth.session-token.0` + `.1` | refresh ~weekly | `error` field of `/api/auth/session` — `doctor` reports "ChatGPT token active". The token is an encrypted JWE, so its `exp` is **not** readable client-side, and the `expires` field is a misleading rolling window; the `error` (`RefreshAccessTokenError` when dead) is the honest signal. |
-| Claude | `sessionKey` | ~30 days | Only detectable via 401 response |
+| Claude | `sessionKey` | ~30 days | Opaque token — only detectable via 401 response |
 
 ### Finding Tokens in Chrome DevTools
 
@@ -102,7 +122,11 @@ Session tokens are how your browser stays logged in. This tool uses them to acce
 4. In the left panel, expand **Cookies** and click the site URL
 5. Find the cookie by name and copy its **Value**
 
-**ChatGPT:** go to `https://chatgpt.com` → find `__Secure-next-auth.session-token` → copy Value (starts with `eyJ`)
+**ChatGPT:** go to `https://chatgpt.com` → find **two** cookies:
+- `__Secure-next-auth.session-token.0` — copy Value (starts with `eyJ`) → `CHATGPT_SESSION_TOKEN`
+- `__Secure-next-auth.session-token.1` — copy Value → `CHATGPT_SESSION_TOKEN_1`
+
+ChatGPT splits large session tokens across two cookies to stay under the browser's 4KB cookie limit. Both are required.
 
 **Claude:** go to `https://claude.ai` → find `sessionKey` → copy Value
 
@@ -138,7 +162,8 @@ cp .env.example .env
 
 | Variable | Description |
 |----------|-------------|
-| `CHATGPT_SESSION_TOKEN` | Your ChatGPT JWT session token (`eyJ…`) |
+| `CHATGPT_SESSION_TOKEN` | ChatGPT session token chunk `.0` (starts with `eyJ…`) |
+| `CHATGPT_SESSION_TOKEN_1` | ChatGPT session token chunk `.1` (the remainder) |
 | `CHATGPT_PROJECT_IDS` | Comma-separated ChatGPT project IDs (see below) |
 | `CLAUDE_SESSION_KEY` | Your Claude session key |
 
@@ -148,6 +173,10 @@ cp .env.example .env
 |----------|---------|-------------|
 | `EXPORT_DIR` | `./exports` | Where to write exported Markdown files |
 | `OUTPUT_STRUCTURE` | `provider/project/year` | Folder structure (see below) |
+| `EXPORTER_HIDDEN_CONTENT` | `placeholder` | What to do with content invisible in the provider's web UI (file-retrieval tool dumps, Custom Instructions): `placeholder` collapses to a one-line note with size, `full` keeps everything, `omit` drops it. Dumps can be 90% of a conversation's bytes. |
+| `EXPORTER_DOWNLOAD_MEDIA` | `images` | Download conversation assets into a `media/` folder beside each export and inline them: `images` (images only), `all` (also audio/voice clips), `off` (text placeholders only). Downloaded media is uploaded to Joplin as resources on the next `joplin` run. |
+| `MAX_CONVERSATIONS_PER_RUN` | unlimited | Cap downloads per export run (per provider). Runs are resumable, so re-running continues where the cap stopped — useful to spread a large first export across several sessions. |
+| `REQUEST_DELAY` | `1.0` | Seconds between consecutive API requests, with small random jitter, so traffic stays human-paced. Set `0` to disable. |
 
 ### Joplin
 
@@ -187,6 +216,19 @@ The `auth` wizard can also guide you through this step interactively.
 
 ---
 
+## Claude Code Sessions
+
+The `claude-code` provider archives your local [Claude Code](https://claude.com/claude-code) agent transcripts — no tokens, no API, no ToS exposure. Sessions are read from `~/.claude/projects/` (override with `CLAUDE_CODE_DIR` in `.env`).
+
+```bash
+ai-chat-exporter export --provider claude-code
+ai-chat-exporter joplin --provider claude-code
+```
+
+Exports are prose-only by default: your prompts and Claude's write-ups are kept, tool activity is grouped into one-line placeholders (`> 🔧 Tool output — 14 calls: Read ×9, Bash ×2 (86KB) — omitted`), and internal reasoning is dropped (counted in the run summary). Set `EXPORTER_HIDDEN_CONTENT=full` to keep everything. Each coding project becomes a Joplin notebook under the `AI-Claude` parent. The provider is included in `--provider all` whenever the sessions directory exists.
+
+---
+
 ## Output Structure
 
 All exported files go under `EXPORT_DIR`. The folder structure maps directly to Joplin notebooks.
@@ -251,10 +293,10 @@ Each provider+project combination maps to a flat Joplin notebook created automat
 ### `auth` — Interactive token setup
 
 ```bash
-ai-chat-exporter auth
+ai-chat-exporter auth                  # manual wizard (DevTools flow)
 ```
 
-Guided wizard to find and save session tokens and ChatGPT project IDs. Detects OS and shows the correct DevTools shortcut.
+Guided wizard to find and save session tokens and ChatGPT project IDs. Detects OS and shows the correct DevTools shortcut, then writes the values to `.env`.
 
 ### `doctor` — Health check
 
@@ -295,7 +337,22 @@ ai-chat-exporter export --output /path/to/my/notes
 ai-chat-exporter export --dry-run
 ```
 
-Options: `--provider [chatgpt|claude|all]`, `--format [markdown|json|both]`, `--output PATH`, `--since YYYY-MM-DD`, `--project NAME`, `--dry-run`
+Options: `--provider [chatgpt|claude|claude-code|all]`, `--format [markdown|json|both]`, `--output PATH`, `--since YYYY-MM-DD`, `--project NAME`, `--hidden-content [full|placeholder|omit]`, `--download-media [images|all|off]`, `--max-conversations N`, `--force`, `--dry-run`
+
+**Re-rendering the whole archive after an upgrade.** New formatting or features (collapse policy, media downloads) only change conversations as they're re-exported. To re-render everything you already have, use `--force` — it re-exports every conversation even if unchanged, **without** `cache --clear`, so your Joplin note links are preserved (a later `joplin` run updates the existing notes instead of duplicating them).
+
+A force re-render runs as a tracked "campaign": each run re-renders the least-recently-exported conversations, the "still to go" count shrinks each run, and finished providers do no further work. The simplest approach is one uncapped pass:
+
+```bash
+ai-chat-exporter export --force      # re-renders everything in one paced pass
+ai-chat-exporter joplin              # update notes + upload media
+```
+
+Or spread the load with `--max-conversations`, re-running until it reports "Force re-render complete":
+
+```bash
+ai-chat-exporter export --force --max-conversations 50   # repeat until complete
+```
 
 ### `list` — List conversations
 
@@ -344,6 +401,22 @@ Reads the local export cache and pushes each exported Markdown file to Joplin as
 
 Options: `--provider [chatgpt|claude|all]`, `--project NAME`, `--dry-run`
 
+### `prune` — Delete stale export files
+
+```bash
+# Preview what would be deleted
+ai-chat-exporter prune --dry-run
+
+# Delete (asks for confirmation; -y skips the prompt)
+ai-chat-exporter prune
+```
+
+Deletes export files no longer referenced by the cache manifest — leftovers
+from old folder layouts or fixed bugs that would otherwise double-sync into
+Joplin. Refuses to run when the manifest is empty (e.g. right after
+`cache --clear`) so it can never wipe a freshly cleared archive. The `doctor`
+command separately verifies that every manifest entry's file exists on disk.
+
 ### `cache` — Manage the sync manifest
 
 ```bash
@@ -421,7 +494,13 @@ Make sure you've added the project IDs to `CHATGPT_PROJECT_IDS` in your `.env`.
 The provider's internal API may have changed. Run with `--debug`, sanitize the output (remove any personal content), and check the project's GitHub Issues for known fixes.
 
 ### Non-text content warnings
-Images, code interpreter outputs, DALL-E generations, and Claude artifacts are not exported in v0.2.0. A WARNING is logged for each skipped item. See `FUTURE.md` for the roadmap.
+Since v0.4.0, rich content is preserved as typed blocks in the export. ChatGPT voice transcripts render as text and audio assets as `📎 File attached` placeholders with size and duration metadata. Anything the extractor doesn't recognise renders as a visible `> ⚠️ Unsupported content` block naming the type and observed keys, *and* increments a counter in the post-export summary so you can tell whether real content is being silently skipped.
+
+### Embedded images and media
+Since v0.6.0, image attachments are downloaded by default into a `media/` folder beside each export and inlined as `![](media/…)`. Audio and other files download only with `EXPORTER_DOWNLOAD_MEDIA=all`. On the next `joplin` run these are uploaded as Joplin resources so they render inside the note. Assets that have expired on the provider's side (older AI-generated images often do) keep their text placeholder and are tallied as `media failed` in the run summary — they're never fatal to the export. Set `EXPORTER_DOWNLOAD_MEDIA=off` to skip downloads entirely.
+
+### Tool output / hidden context collapsed
+Since v0.6.0, content that was invisible in the ChatGPT web UI is collapsed by default to one-line placeholders like `> 🔧 Tool output — file_search (15.0 KB) — omitted`. This is ChatGPT's file-retrieval tool re-injecting your attached files on every run — it can be 90% of a conversation's bytes while containing none of the dialogue. Custom Instructions collapse the same way (`> ℹ️ Hidden context`). The post-export summary lists every collapsed message by origin with the total KB omitted. To keep everything, set `EXPORTER_HIDDEN_CONTENT=full` in `.env` or pass `--hidden-content full`.
 
 ### Empty export / all conversations skipped
 No new or updated conversations since your last run. To verify: `ai-chat-exporter cache --show`. To force a full re-export: `ai-chat-exporter cache --clear`.
@@ -435,12 +514,9 @@ No new or updated conversations since your last run. To verify: `ai-chat-exporte
 
 ## Future Work
 
-See `FUTURE.md` for planned features:
+See `FUTURE.md` for the full roadmap. Current priorities:
 
-- **v0.2.x** — `export --force` flag; `joplin --force` flag; per-conversation cache reset
+- **Watch/scheduled mode** on the way to a headless StartOS service
-- **v0.3.0** — Official API fallback: parse export ZIP files from ChatGPT/Claude settings
-- **v0.4.0** — Rich content: images, artifacts, code interpreter output, extended thinking
-- **v0.5.0** — Watch/scheduled mode; Obsidian vault output
 
 ---
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ai-chat-exporter"
-version = "0.2.1"
+version = "0.7.0"
 description = "Export ChatGPT and Claude conversation history to Markdown for personal archival in Joplin"
 requires-python = ">=3.11"
 dependencies = [

src/blocks.py

@@ -0,0 +1,399 @@
+"""Typed content blocks for normalized messages.
+
+Providers produce ordered lists of blocks; exporters render them. Living outside
+``src/providers/`` deliberately — blocks are a separate concern from extraction
+or rendering, shared by both layers.
+
+Block dicts always have ``type`` set to one of the BLOCK_TYPE_* constants.
+Construct via the ``make_*`` helpers; never build dicts by hand. The ``unknown``
+block constructor REQUIRES a corresponding WARNING log + ``LossReport`` tally
+at the call site — see plan §Data-loss visibility.
+"""
+
+import json
+from pathlib import Path
+from typing import Any
+
+BLOCK_TYPE_TEXT = "text"
+BLOCK_TYPE_CODE = "code"
+BLOCK_TYPE_THINKING = "thinking"
+BLOCK_TYPE_TOOL_USE = "tool_use"
+BLOCK_TYPE_TOOL_RESULT = "tool_result"
+BLOCK_TYPE_CITATION = "citation"
+BLOCK_TYPE_IMAGE_PLACEHOLDER = "image_placeholder"
+BLOCK_TYPE_FILE_PLACEHOLDER = "file_placeholder"
+BLOCK_TYPE_UNKNOWN = "unknown"
+BLOCK_TYPE_HIDDEN_CONTEXT_MARKER = "hidden_context_marker"
+BLOCK_TYPE_COLLAPSED = "collapsed"
+
+COLLAPSED_KIND_TOOL_DUMP = "tool_dump"
+COLLAPSED_KIND_HIDDEN_CONTEXT = "hidden_context"
+
+UNKNOWN_REASON_UNKNOWN_TYPE = "unknown_type"
+UNKNOWN_REASON_EXTRACTION_FAILED = "extraction_failed"
+UNKNOWN_REASON_ALL_BLOCKS_FAILED = "all_blocks_failed"
+UNKNOWN_REASON_UNKNOWN_FIELD_IN_KNOWN_TYPE = "unknown_field_in_known_type"
+
+_OBSERVED_KEYS_LIMIT = 10
+
+
+# ---------------------------------------------------------------------------
+# Constructors
+# ---------------------------------------------------------------------------
+
+
+def make_text_block(text: str) -> dict | None:
+    """Return a text block, or None if the text is empty/whitespace-only.
+
+    Returning None lets callers do ``if block: blocks.append(block)`` and prune
+    empty blocks at construction time. See plan §Finalizing the message dict.
+    """
+    if not isinstance(text, str) or not text.strip():
+        return None
+    return {"type": BLOCK_TYPE_TEXT, "text": text}
+
+
+def make_code_block(code: str, language: str = "") -> dict | None:
+    """Return a code block, or None if code is empty."""
+    if not isinstance(code, str) or not code.strip():
+        return None
+    return {"type": BLOCK_TYPE_CODE, "language": language or "", "code": code}
+
+
+def make_thinking_block(text: str) -> dict | None:
+    """Return a thinking block, or None if empty."""
+    if not isinstance(text, str) or not text.strip():
+        return None
+    return {"type": BLOCK_TYPE_THINKING, "text": text}
+
+
+def make_tool_use_block(name: str, input_data: Any, tool_id: str | None = None) -> dict:
+    """Return a tool_use block.
+
+    Always returns a block (no None) — tool calls are meaningful even with
+    empty inputs.
+    """
+    return {
+        "type": BLOCK_TYPE_TOOL_USE,
+        "name": name or "",
+        "input": input_data if input_data is not None else {},
+        "tool_id": tool_id,
+    }
+
+
+def make_tool_result_block(
+    output: str,
+    tool_name: str | None = None,
+    is_error: bool = False,
+    summary: str | None = None,
+) -> dict:
+    """Return a tool_result block.
+
+    ``summary`` is an optional short human label rendered between header and
+    fence (e.g. ChatGPT's ``metadata.reasoning_title`` for execution_output).
+    """
+    return {
+        "type": BLOCK_TYPE_TOOL_RESULT,
+        "tool_name": tool_name,
+        "output": output if isinstance(output, str) else str(output),
+        "is_error": bool(is_error),
+        "summary": summary,
+    }
+
+
+def make_citation_block(
+    url: str,
+    title: str | None = None,
+    snippet: str | None = None,
+) -> dict | None:
+    if not url:
+        return None
+    return {
+        "type": BLOCK_TYPE_CITATION,
+        "url": url,
+        "title": title,
+        "snippet": snippet,
+    }
+
+
+def make_image_placeholder(
+    ref: str,
+    source: str = "unknown",
+    mime: str | None = None,
+) -> dict:
+    """source ∈ {'user_upload', 'model_generated', 'unknown'}."""
+    return {
+        "type": BLOCK_TYPE_IMAGE_PLACEHOLDER,
+        "ref": ref or "",
+        "source": source,
+        "mime": mime,
+    }
+
+
+def make_file_placeholder(
+    ref: str,
+    filename: str | None = None,
+    mime: str | None = None,
+    size_bytes: int | None = None,
+    duration_seconds: float | None = None,
+) -> dict:
+    return {
+        "type": BLOCK_TYPE_FILE_PLACEHOLDER,
+        "ref": ref or "",
+        "filename": filename,
+        "mime": mime,
+        "size_bytes": size_bytes,
+        "duration_seconds": duration_seconds,
+    }
+
+
+def make_unknown_block(
+    raw_type: str,
+    observed_keys: list[str] | None = None,
+    reason: str = UNKNOWN_REASON_UNKNOWN_TYPE,
+    summary: str | None = None,
+) -> dict:
+    """Construct an unknown block.
+
+    Every call site MUST also emit a WARNING log and increment a LossReport
+    tally — see plan §Data-loss visibility. The block surfaces the loss at
+    read time; the WARNING surfaces it at run time. Both signals matter.
+    """
+    keys = list(observed_keys or [])[:_OBSERVED_KEYS_LIMIT]
+    return {
+        "type": BLOCK_TYPE_UNKNOWN,
+        "raw_type": raw_type,
+        "observed_keys": keys,
+        "reason": reason,
+        "summary": summary,
+    }
+
+
+def make_collapsed_block(
+    origin: str,
+    content_type: str,
+    size_bytes: int,
+    kind: str,
+) -> dict:
+    """One-line stand-in for a message omitted by the EXPORTER_HIDDEN_CONTENT policy.
+
+    ``kind`` ∈ {COLLAPSED_KIND_TOOL_DUMP, COLLAPSED_KIND_HIDDEN_CONTEXT}.
+    Unlike ``unknown`` blocks (unexpected loss), a collapsed block records an
+    intentional policy decision — the content was invisible in the provider's
+    web UI and the user chose not to keep it. Every call site MUST tally via
+    ``LossReport.record_collapsed`` so the omission stays visible in the
+    post-export summary.
+    """
+    return {
+        "type": BLOCK_TYPE_COLLAPSED,
+        "origin": origin or "",
+        "content_type": content_type or "",
+        "size_bytes": int(size_bytes or 0),
+        "kind": kind,
+    }
+
+
+def make_hidden_context_marker(content_type: str) -> dict:
+    """A short prepend block that flags the surrounding message as hidden context.
+
+    Driven by the ``metadata.is_visually_hidden_from_conversation`` flag, not by
+    content_type matching. The marker tells the reader "this message is
+    hidden in the source UI; we're showing it here for archival fidelity."
+    """
+    return {
+        "type": BLOCK_TYPE_HIDDEN_CONTEXT_MARKER,
+        "content_type": content_type or "",
+    }
+
+
+# ---------------------------------------------------------------------------
+# Rendering
+# ---------------------------------------------------------------------------
+
+
+def render_blocks_to_markdown(blocks: list[dict]) -> str:
+    """Render an ordered list of blocks to a single Markdown string.
+
+    Blocks are joined with one blank line between them. Pure function; no I/O.
+    """
+    if not blocks:
+        return ""
+
+    rendered: list[str] = []
+    for block in blocks:
+        chunk = _render_one(block)
+        if chunk:
+            rendered.append(chunk)
+
+    return "\n\n".join(rendered)
+
+
+def _render_one(block: dict) -> str:
+    btype = block.get("type", "")
+    if btype == BLOCK_TYPE_TEXT:
+        return block.get("text", "")
+    if btype == BLOCK_TYPE_CODE:
+        lang = block.get("language") or ""
+        code = block.get("code", "")
+        fence = _safe_fence(code)
+        return f"{fence}{lang}\n{code}\n{fence}"
+    if btype == BLOCK_TYPE_THINKING:
+        text = block.get("text", "")
+        quoted = _blockquote_prefix(text)
+        return f"**💭 Reasoning**\n\n{quoted}"
+    if btype == BLOCK_TYPE_TOOL_USE:
+        name = block.get("name", "")
+        input_data = block.get("input", {})
+        body_json = json.dumps(input_data, indent=2, sort_keys=False, default=str, ensure_ascii=False)
+        fence = _safe_fence(body_json)
+        body = f"{fence}json\n{body_json}\n{fence}"
+        quoted = _blockquote_prefix(f"🔧 **Tool: {name}**\n{body}")
+        return quoted
+    if btype == BLOCK_TYPE_TOOL_RESULT:
+        output = block.get("output", "")
+        is_error = bool(block.get("is_error"))
+        tool_name = block.get("tool_name") or ""
+        summary = block.get("summary") or ""
+        icon = "❌" if is_error else "📤"
+        label = "Result (error)" if is_error else "Result"
+        if tool_name:
+            header = f"{icon} **{label}: {tool_name}**"
+        else:
+            header = f"{icon} **{label}**"
+        fence = _safe_fence(output)
+        body = f"{fence}\n{output}\n{fence}"
+        if summary:
+            inner = f"{header}\n*{summary}*\n{body}"
+        else:
+            inner = f"{header}\n{body}"
+        return _blockquote_prefix(inner)
+    if btype == BLOCK_TYPE_CITATION:
+        url = block.get("url", "")
+        title = block.get("title") or url
+        return f"[{title}]({url})"
+    if btype == BLOCK_TYPE_IMAGE_PLACEHOLDER:
+        ref = block.get("ref", "")
+        source = block.get("source", "unknown")
+        mime = block.get("mime")
+        local_path = block.get("local_path")
+        if local_path:
+            # Downloaded — render as a real inline image.
+            return f"![{source or 'image'}]({local_path})"
+        meta_parts = [source] if source else []
+        if mime:
+            meta_parts.append(mime)
+        meta_parts.append("content not preserved in this export")
+        meta = ", ".join(meta_parts)
+        return f"> 🖼️ **Image attached** — `{ref}` ({meta})"
+    if btype == BLOCK_TYPE_FILE_PLACEHOLDER:
+        ref = block.get("ref", "")
+        filename = block.get("filename")
+        label = filename or ref
+        mime = block.get("mime")
+        size_bytes = block.get("size_bytes")
+        duration = block.get("duration_seconds")
+        local_path = block.get("local_path")
+        meta_parts: list[str] = []
+        if mime:
+            meta_parts.append(mime)
+        size_label = _format_size(size_bytes)
+        if size_label:
+            meta_parts.append(size_label)
+        if isinstance(duration, (int, float)) and duration > 0:
+            meta_parts.append(f"{duration:.2f}s")
+        if local_path:
+            # Downloaded — render as a link to the local copy.
+            meta = ", ".join(meta_parts) if meta_parts else ""
+            meta_str = f" ({meta})" if meta else ""
+            return f"> 📎 **File attached** — [{Path(local_path).name}]({local_path}){meta_str}"
+        meta_parts.append("content not preserved in this export")
+        meta = ", ".join(meta_parts)
+        return f"> 📎 **File attached** — `{label}` ({meta})"
+    if btype == BLOCK_TYPE_UNKNOWN:
+        raw_type = block.get("raw_type", "?")
+        reason = block.get("reason", UNKNOWN_REASON_UNKNOWN_TYPE)
+        keys = block.get("observed_keys") or []
+        summary = block.get("summary")
+        first_line = f"⚠️ **Unsupported content** — type `{raw_type}` ({reason})"
+        lines = [first_line]
+        if summary:
+            lines.append(summary)
+        if keys:
+            keys_str = ", ".join(f"`{k}`" for k in keys)
+            lines.append(f"Keys observed: {keys_str}")
+        return _blockquote_prefix("\n".join(lines))
+    if btype == BLOCK_TYPE_HIDDEN_CONTEXT_MARKER:
+        ctype = block.get("content_type", "")
+        return f"> ℹ️ **Hidden context** — `{ctype}`"
+    if btype == BLOCK_TYPE_COLLAPSED:
+        origin = block.get("origin", "")
+        kind = block.get("kind", "")
+        if kind == COLLAPSED_KIND_HIDDEN_CONTEXT:
+            icon, label = "ℹ️", "Hidden context"
+        else:
+            icon, label = "🔧", "Tool output"
+        size_label = _format_size(block.get("size_bytes"))
+        size_part = f" ({size_label})" if size_label else ""
+        return (
+            f"> {icon} **{label}** — `{origin}`{size_part} — omitted "
+            "(EXPORTER_HIDDEN_CONTENT=full to keep)"
+        )
+
+    # Defensive: a block of unrecognised local type (shouldn't happen if
+    # constructors are used). Render as visible warning rather than dropping.
+    return f"> ⚠️ **Internal: unrecognised block type** — `{btype}`"
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _format_size(size_bytes: Any) -> str:
+    """Human-readable size ('24.1 KB', '1.05 MB'), or '' when absent/zero."""
+    if not isinstance(size_bytes, int) or size_bytes <= 0:
+        return ""
+    kb = size_bytes / 1024
+    return f"{kb:.1f} KB" if kb < 1024 else f"{kb / 1024:.2f} MB"
+
+
+def _safe_fence(text: str) -> str:
+    """Return a backtick fence longer than the longest run of backticks in ``text``.
+
+    CommonMark requires the closing fence to be at least as long as the opening
+    fence. Picking N+1 (where N = longest run in content) ensures the content's
+    own backticks are inert. Minimum is 3.
+
+    Verified live against Joplin during planning — see plan
+    §Backtick-corruption defense.
+    """
+    if not isinstance(text, str):
+        return "```"
+    longest_run = 0
+    current_run = 0
+    for ch in text:
+        if ch == "`":
+            current_run += 1
+            if current_run > longest_run:
+                longest_run = current_run
+        else:
+            current_run = 0
+    fence_len = max(3, longest_run + 1)
+    return "`" * fence_len
+
+
+def _blockquote_prefix(text: str) -> str:
+    """Prefix every line of ``text`` with ``> `` so the whole block renders as a quote.
+
+    Empty source lines become ``>`` (no trailing space) so blockquote continuity
+    is preserved without trailing-whitespace noise.
+    """
+    if not isinstance(text, str):
+        return ""
+    out_lines: list[str] = []
+    for line in text.split("\n"):
+        if line == "":
+            out_lines.append(">")
+        else:
+            out_lines.append(f"> {line}")
+    return "\n".join(out_lines)

src/cache.py

@@ -84,27 +84,70 @@ class Cache:
         if provider not in self._data:
             self._data[provider] = {}
 
-        self._data[provider][conv_id] = {
+        # Preserve the Joplin link across re-exports. Without this, re-rendering
+        # a conversation (e.g. a forced re-export to pick up new formatting)
+        # would drop joplin_note_id, and the next sync would create a duplicate
+        # note instead of updating the existing one. exported_at is refreshed,
+        # so get_joplin_pending() still flags the note for an update.
+        prev = self._data[provider].get(conv_id) or {}
+        entry = {
             "title": metadata.get("title", ""),
             "project": metadata.get("project"),
+            "created_at": metadata.get("created_at", ""),
             "updated_at": metadata.get("updated_at", ""),
             "exported_at": datetime.now(tz=timezone.utc).isoformat(),
             "file_path": metadata.get("file_path", ""),
         }
+        for carried in ("joplin_note_id", "joplin_synced_at", "joplin_resources"):
+            if carried in prev:
+                entry[carried] = prev[carried]
+
+        self._data[provider][conv_id] = entry
         self._data["last_run"] = datetime.now(tz=timezone.utc).isoformat()
         self._save()
 
-    def get_new_or_updated(self, provider: str, conversations: list[dict]) -> list[dict]:
+    def get_new_or_updated(
+        self,
+        provider: str,
+        conversations: list[dict],
+        force: bool = False,
+        campaign_at: str | None = None,
+    ) -> list[dict]:
         """Filter a conversation list to only new or updated conversations.
 
         Args:
             provider: "chatgpt" or "claude"
             conversations: List of raw conversation dicts from the provider.
                            Each must have an ``id``/``uuid`` and ``updated_at``/``update_time``.
+            force: When True, return conversations to re-render regardless of
+                   cache freshness — used by ``export --force``.
+            campaign_at: In force mode, exclude conversations already re-exported
+                   at/after this ISO timestamp (the re-render campaign start), so
+                   each run advances and finished providers do nothing.
 
         Returns:
-            Subset that needs to be exported.
+            Subset that needs to be exported, ordered oldest-export-first in
+            force mode.
         """
+        if force:
+            # Re-export for a re-render campaign, ordered least-recently-exported
+            # first (never-exported sorts first). When ``campaign_at`` is given,
+            # conversations already re-exported during this campaign
+            # (exported_at >= campaign_at) are excluded — so finished providers
+            # do no work and the remaining count shrinks to zero. Each capped run
+            # re-renders the next-oldest batch. Conversations without an id drop.
+            entries = self._data.get(provider, {})
+
+            def _exported_at(conv: dict) -> str:
+                conv_id = conv.get("id") or conv.get("uuid", "")
+                entry = entries.get(conv_id) or {}
+                return entry.get("exported_at") or ""
+
+            candidates = [c for c in conversations if c.get("id") or c.get("uuid")]
+            if campaign_at is not None:
+                candidates = [c for c in candidates if _exported_at(c) < campaign_at]
+            return sorted(candidates, key=_exported_at)
+
         result = []
         for conv in conversations:
             conv_id = conv.get("id") or conv.get("uuid", "")
@@ -172,6 +215,22 @@ class Cache:
         entry["joplin_synced_at"] = datetime.now(tz=timezone.utc).isoformat()
         self._save()
 
+    def get_joplin_resources(self, provider: str, conv_id: str) -> dict[str, str]:
+        """Return the {relative_media_path: joplin_resource_id} map for a conversation."""
+        entry = self._data.get(provider, {}).get(conv_id)
+        if not isinstance(entry, dict):
+            return {}
+        resources = entry.get("joplin_resources")
+        return dict(resources) if isinstance(resources, dict) else {}
+
+    def set_joplin_resources(self, provider: str, conv_id: str, resources: dict[str, str]) -> None:
+        """Persist the {relative_media_path: joplin_resource_id} map for a conversation."""
+        entry = self._data.get(provider, {}).get(conv_id)
+        if entry is None:
+            return
+        entry["joplin_resources"] = dict(resources)
+        self._save()
+
     def get_joplin_pending(self, provider: str) -> list[tuple[str, dict]]:
         """Return (conv_id, entry) pairs that need to be synced to Joplin.
 
@@ -209,10 +268,48 @@ class Cache:
 
         return pending
 
+    def all_file_paths(self) -> set[str]:
+        """Return every non-empty ``file_path`` in the manifest, across providers.
+
+        Used by ``prune`` (files on disk but not here are stale) and by the
+        doctor integrity check (files here but not on disk are missing).
+        """
+        paths: set[str] = set()
+        for key, entries in self._data.items():
+            if not isinstance(entries, dict) or key in (
+                "version",
+                "last_run",
+                "tos_acknowledged_at",
+            ):
+                continue
+            for entry in entries.values():
+                if isinstance(entry, dict) and entry.get("file_path"):
+                    paths.add(entry["file_path"])
+        return paths
+
     def last_run(self) -> str | None:
         """Return the ISO8601 timestamp of the last export run, or None."""
         return self._data.get("last_run")
 
+    def exported_at(self, provider: str, conv_id: str) -> str:
+        """Return the recorded ``exported_at`` for a conversation, or '' if absent."""
+        entry = self._data.get(provider, {}).get(conv_id)
+        return entry.get("exported_at", "") if isinstance(entry, dict) else ""
+
+    # Force re-render campaign marker (top-level scalar; skipped by stats/clear,
+    # which only act on dict-valued provider keys).
+    def get_force_campaign(self) -> str | None:
+        """Return the active force-re-render campaign start timestamp, or None."""
+        return self._data.get("force_campaign_at")
+
+    def set_force_campaign(self, started_at: str) -> None:
+        self._data["force_campaign_at"] = started_at
+        self._save()
+
+    def clear_force_campaign(self) -> None:
+        if self._data.pop("force_campaign_at", None) is not None:
+            self._save()
+
     # ------------------------------------------------------------------
     # Private helpers
     # ------------------------------------------------------------------

src/config.py

@@ -20,6 +20,12 @@ _CLAUDE_PLACEHOLDER = ""
 # Valid OUTPUT_STRUCTURE values
 VALID_STRUCTURES = {"provider/project/year", "provider/project", "provider/year"}
 
+# Valid EXPORTER_HIDDEN_CONTENT values (see src/providers/base.py)
+VALID_HIDDEN_CONTENT = {"full", "placeholder", "omit"}
+
+# Valid EXPORTER_DOWNLOAD_MEDIA values (see src/media.py)
+VALID_DOWNLOAD_MEDIA = {"images", "all", "off"}
+
 
 class ConfigError(Exception):
     """Raised when required configuration is missing or invalid."""
@@ -43,6 +49,16 @@ class Config:
     # Joplin local REST API settings (Web Clipper service)
     joplin_api_token: str | None = None
     joplin_api_url: str = "http://localhost:41184"
+    # Policy for content invisible in the provider web UI (retrieval dumps,
+    # hidden context): full | placeholder | omit
+    hidden_content: str = "placeholder"
+    # Session cap: max conversations downloaded per export run (None = unlimited).
+    # Every run is resumable, so a capped run just continues next time.
+    max_conversations: int | None = None
+    # Seconds between consecutive API requests (politeness pacing; 0 disables)
+    request_delay: float = 1.0
+    # Asset download policy: images | all | off
+    download_media: str = "images"
 
 
 def load_config() -> Config:
@@ -67,6 +83,12 @@ def load_config() -> Config:
     joplin_token = os.getenv("JOPLIN_API_TOKEN", "").strip() or None
     joplin_url = os.getenv("JOPLIN_API_URL", "http://localhost:41184").strip()
 
+    hidden_content = os.getenv("EXPORTER_HIDDEN_CONTENT", "").strip().lower() or "placeholder"
+
+    max_conversations_raw = os.getenv("MAX_CONVERSATIONS_PER_RUN", "").strip()
+    request_delay_raw = os.getenv("REQUEST_DELAY", "").strip()
+    download_media = os.getenv("EXPORTER_DOWNLOAD_MEDIA", "").strip().lower() or "images"
+
     # Parse CHATGPT_PROJECT_IDS — comma-separated list of gizmo IDs (g-p-xxx)
     _project_ids_raw = os.getenv("CHATGPT_PROJECT_IDS", "").strip()
     chatgpt_project_ids = [
@@ -90,6 +112,46 @@ def load_config() -> Config:
             f"Must be one of: {', '.join(sorted(VALID_STRUCTURES))}"
         )
 
+    # Validate hidden-content policy
+    if hidden_content not in VALID_HIDDEN_CONTENT:
+        errors.append(
+            f"EXPORTER_HIDDEN_CONTENT '{hidden_content}' is invalid. "
+            f"Must be one of: {', '.join(sorted(VALID_HIDDEN_CONTENT))}"
+        )
+
+    # Validate media download policy
+    if download_media not in VALID_DOWNLOAD_MEDIA:
+        errors.append(
+            f"EXPORTER_DOWNLOAD_MEDIA '{download_media}' is invalid. "
+            f"Must be one of: {', '.join(sorted(VALID_DOWNLOAD_MEDIA))}"
+        )
+
+    # Validate session cap
+    max_conversations: int | None = None
+    if max_conversations_raw:
+        try:
+            max_conversations = int(max_conversations_raw)
+        except ValueError:
+            errors.append(
+                f"MAX_CONVERSATIONS_PER_RUN '{max_conversations_raw}' is not an integer."
+            )
+        else:
+            if max_conversations < 1:
+                errors.append(
+                    f"MAX_CONVERSATIONS_PER_RUN must be at least 1 (got {max_conversations})."
+                )
+
+    # Validate request pacing
+    request_delay = 1.0
+    if request_delay_raw:
+        try:
+            request_delay = float(request_delay_raw)
+        except ValueError:
+            errors.append(f"REQUEST_DELAY '{request_delay_raw}' is not a number.")
+        else:
+            if request_delay < 0:
+                errors.append(f"REQUEST_DELAY must be >= 0 (got {request_delay}).")
+
     # Validate and decode ChatGPT JWT
     chatgpt_expiry: datetime | None = None
     if chatgpt_token:
@@ -139,6 +201,10 @@ def load_config() -> Config:
         chatgpt_project_ids=chatgpt_project_ids,
         joplin_api_token=joplin_token,
         joplin_api_url=joplin_url,
+        hidden_content=hidden_content,
+        max_conversations=max_conversations,
+        request_delay=request_delay,
+        download_media=download_media,
     )
 
     _log_startup_summary(config)
@@ -223,7 +289,11 @@ def _log_startup_summary(cfg: Config) -> None:
         "Joplin: %s | "
         "export_dir=%s | "
         "structure=%s | "
-        "cache_dir=%s",
+        "cache_dir=%s | "
+        "hidden_content=%s | "
+        "max_conversations=%s | "
+        "request_delay=%.1fs | "
+        "download_media=%s",
         chatgpt_status,
         claude_status,
         len(cfg.chatgpt_project_ids),
@@ -231,4 +301,8 @@ def _log_startup_summary(cfg: Config) -> None:
         cfg.export_dir,
         cfg.output_structure,
         cfg.cache_dir,
+        cfg.hidden_content,
+        cfg.max_conversations if cfg.max_conversations is not None else "unlimited",
+        cfg.request_delay,
+        cfg.download_media,
     )

src/exporters/markdown.py

@@ -6,6 +6,7 @@ from datetime import datetime, timezone
 from pathlib import Path
 from typing import Any
 
+from src.blocks import render_blocks_to_markdown
 from src.utils import build_export_path, generate_filename
 
 logger = logging.getLogger(__name__)
@@ -15,6 +16,7 @@ _ROLE_LABELS = {
     "user": ("🧑 Human", "user"),
     "assistant": ("🤖 Assistant", "assistant"),
     "system": ("⚙️ System", "system"),
+    "tool": ("🔧 Tool", "tool"),
 }
 
 
@@ -125,10 +127,17 @@ class MarkdownExporter:
         # Messages
         for msg in messages:
             role = msg.get("role", "user")
-            content = msg.get("content", "")
+            blocks = msg.get("blocks") or []
             timestamp = msg.get("timestamp")
 
-            if not content or not content.strip():
+            # Prefer rendering from blocks (v0.4.0+). Backward-compat fallback:
+            # if blocks is missing/empty AND content exists, render content as-is.
+            if blocks:
+                body = render_blocks_to_markdown(blocks)
+            else:
+                body = msg.get("content", "") or ""
+
+            if not body or not body.strip():
                 logger.warning(
                     "[markdown] Skipping empty/whitespace message in conversation %s",
                     conv_id[:8],
@@ -143,7 +152,7 @@ class MarkdownExporter:
             else:
                 lines.append("")
 
-            lines.append(content)
+            lines.append(body)
             lines.append("")
             lines.append("---")
             lines.append("")

src/joplin.py

@@ -1,7 +1,10 @@
 """Joplin Data API client for importing notes into Joplin desktop."""
 
+import json
 import logging
 import os
+import re
+from pathlib import Path
 from typing import Any
 
 import requests
@@ -32,8 +35,8 @@ class JoplinClient:
     def __init__(self, base_url: str, token: str) -> None:
         self._base_url = base_url.rstrip("/")
         self._token = token
-        # In-memory cache of notebook title → ID to avoid repeated GET /folders
+        # In-memory cache: (parent_id | None, title) → folder ID
-        self._notebook_cache: dict[str, str] = {}
+        self._notebook_cache: dict[tuple[str | None, str], str] = {}
         self._notebooks_loaded = False
         logger.debug("[joplin] Client initialised with base_url=%s", self._base_url)
 
@@ -89,13 +92,13 @@ class JoplinClient:
         """Return all Joplin notebooks (folders), handling pagination.
 
         Returns:
-            List of folder dicts with at least ``id`` and ``title`` keys.
+            List of folder dicts with at least ``id``, ``title``, and ``parent_id`` keys.
         """
         results: list[dict] = []
         page = 1
         while True:
             logger.debug("[joplin] GET /folders page=%d", page)
-            resp = self._get("/folders", params={"page": page, "fields": "id,title"})
+            resp = self._get("/folders", params={"page": page, "fields": "id,title,parent_id"})
             items = resp.get("items", [])
             results.extend(items)
             logger.debug("[joplin] /folders page=%d → %d items, has_more=%s", page, len(items), resp.get("has_more"))
@@ -104,11 +107,12 @@ class JoplinClient:
             page += 1
         return results
 
-    def get_or_create_notebook(self, title: str) -> str:
+    def get_or_create_notebook(self, title: str, parent_id: str | None = None) -> str:
-        """Return the Joplin folder ID for ``title``, creating it if needed.
+        """Return the Joplin folder ID for ``title`` under ``parent_id``, creating if needed.
 
         Args:
-            title: Notebook display name (e.g. "ChatGPT - My Project").
+            title: Notebook display name.
+            parent_id: ID of the parent folder, or None for a root notebook.
 
         Returns:
             Joplin folder ID string.
@@ -116,19 +120,40 @@ class JoplinClient:
         if not self._notebooks_loaded:
             self._load_notebook_cache()
 
-        if title in self._notebook_cache:
+        key = (parent_id, title)
-            folder_id = self._notebook_cache[title]
+        if key in self._notebook_cache:
-            logger.debug("[joplin] Notebook cache hit: %r → %s", title, folder_id)
+            folder_id = self._notebook_cache[key]
+            logger.debug("[joplin] Notebook cache hit: %r (parent=%s) → %s", title, parent_id, folder_id)
             return folder_id
 
         # Not found — create it
-        logger.info("[joplin] Creating notebook: %r", title)
+        logger.info("[joplin] Creating notebook: %r (parent=%s)", title, parent_id)
-        resp = self._post("/folders", {"title": title})
+        data: dict = {"title": title}
+        if parent_id:
+            data["parent_id"] = parent_id
+        resp = self._post("/folders", data)
         folder_id = resp["id"]
-        self._notebook_cache[title] = folder_id
+        self._notebook_cache[key] = folder_id
         logger.debug("[joplin] Notebook created: %r → %s", title, folder_id)
         return folder_id
 
+    def get_or_create_notebook_path(self, path: list[str]) -> str:
+        """Ensure a nested notebook path exists and return the leaf folder ID.
+
+        Creates intermediate notebooks as needed.
+
+        Args:
+            path: Ordered list of notebook names, e.g. ["AI-ChatGPT", "No Project"].
+
+        Returns:
+            Joplin folder ID of the deepest (leaf) notebook.
+        """
+        parent_id: str | None = None
+        for name in path:
+            parent_id = self.get_or_create_notebook(name, parent_id)
+        assert parent_id is not None
+        return parent_id
+
     # ------------------------------------------------------------------
     # Notes
     # ------------------------------------------------------------------
@@ -168,6 +193,46 @@ class JoplinClient:
         self._put(f"/notes/{note_id}", {"title": title, "body": body})
         logger.info("[joplin] Note updated: %r (%s)", title, note_id)
 
+    # ------------------------------------------------------------------
+    # Resources (embedded media)
+    # ------------------------------------------------------------------
+
+    def create_resource(self, file_path: "Path", title: str | None = None) -> str:
+        """Upload a local file as a Joplin resource and return its ID.
+
+        Joplin's ``POST /resources`` is multipart: a ``data`` file part plus a
+        ``props`` JSON part. The returned ID is referenced from note bodies as
+        ``:/<id>``.
+        """
+        url = f"{self._base_url}/resources"
+        props = json.dumps({"title": title or file_path.name})
+        logger.debug("[joplin] POST /resources (%s)", file_path.name)
+        try:
+            with file_path.open("rb") as fh:
+                resp = requests.post(
+                    url,
+                    params={"token": self._token},
+                    files={
+                        "data": (file_path.name, fh),
+                        "props": (None, props),
+                    },
+                    timeout=_REQUEST_TIMEOUT,
+                )
+            resp.raise_for_status()
+            resource_id = resp.json()["id"]
+            logger.info("[joplin] Resource created: %s → %s", file_path.name, resource_id)
+            return resource_id
+        except requests.exceptions.ConnectionError as e:
+            raise JoplinError(
+                "Cannot connect to Joplin. Is Joplin desktop running with Web Clipper enabled?"
+            ) from e
+        except requests.exceptions.Timeout as e:
+            raise JoplinError(_timeout_message("POST", "/resources")) from e
+        except requests.exceptions.HTTPError as e:
+            raise JoplinError(_http_error_message("POST", "/resources", e)) from e
+        except (requests.exceptions.RequestException, KeyError, OSError) as e:
+            raise JoplinError(f"Joplin resource upload failed for {file_path.name}: {e}") from e
+
     # ------------------------------------------------------------------
     # HTTP helpers
     # ------------------------------------------------------------------
@@ -233,11 +298,14 @@ class JoplinClient:
     def _load_notebook_cache(self) -> None:
         logger.debug("[joplin] Loading notebook list from Joplin…")
         notebooks = self.list_notebooks()
-        self._notebook_cache = {nb["title"]: nb["id"] for nb in notebooks}
+        self._notebook_cache = {
+            (nb.get("parent_id") or None, nb["title"]): nb["id"]
+            for nb in notebooks
+        }
         self._notebooks_loaded = True
         logger.debug("[joplin] Notebook cache loaded: %d notebooks", len(self._notebook_cache))
-        for title, folder_id in self._notebook_cache.items():
+        for (parent_id, title), folder_id in self._notebook_cache.items():
-            logger.debug("[joplin]   %r → %s", title, folder_id)
+            logger.debug("[joplin]   (%s) %r → %s", parent_id or "root", title, folder_id)
 
 
 # ------------------------------------------------------------------
@@ -279,25 +347,70 @@ def _http_error_message(method: str, path: str, e: requests.exceptions.HTTPError
     return f"Joplin {method} {path} failed: HTTP {status}{body_snippet}"
 
 
+# ------------------------------------------------------------------
+# Embedded-media rewriting
+# ------------------------------------------------------------------
+
+# Markdown image/link targets pointing at the local media/ sibling directory,
+# e.g. ![alt](media/file_x.png) or [name](media/clip.wav).
+_MEDIA_LINK_RE = re.compile(r"(!?\[[^\]]*\])\((media/[^)\s]+)\)")
+
+
+def upload_media_and_rewrite(
+    body: str,
+    note_dir: Path,
+    client: "JoplinClient",
+    resource_map: dict[str, str],
+) -> tuple[str, dict[str, str]]:
+    """Rewrite local ``media/…`` links to Joplin ``:/resourceId`` references.
+
+    Uploads each referenced file as a Joplin resource (once), rewriting both
+    image embeds and file links. ``resource_map`` (relative-path → resource_id)
+    is the idempotency record from the cache: known paths are reused, new ones
+    are added and returned so the caller can persist them. Missing files are
+    left as-is so the link still points at the on-disk copy.
+    """
+    new_map = dict(resource_map)
+
+    def replace(match: re.Match) -> str:
+        label, rel_path = match.group(1), match.group(2)
+        resource_id = new_map.get(rel_path)
+        if resource_id is None:
+            file_path = (note_dir / rel_path).resolve()
+            if not file_path.is_file():
+                logger.warning("[joplin] Media file missing, leaving link: %s", rel_path)
+                return match.group(0)
+            resource_id = client.create_resource(file_path)
+            new_map[rel_path] = resource_id
+        return f"{label}(:/{resource_id})"
+
+    return _MEDIA_LINK_RE.sub(replace, body), new_map
+
+
 # ------------------------------------------------------------------
 # Notebook naming helper
 # ------------------------------------------------------------------
 
 
 _PROVIDER_DISPLAY = {
-    "chatgpt": "ChatGPT",
+    "chatgpt": "AI-ChatGPT",
-    "claude": "Claude",
+    "claude": "AI-Claude",
+    # Decision 2026-06-12: Claude Code coding projects nest under the same
+    # AI-Claude parent, alongside Claude web projects.
+    "claude-code": "AI-Claude",
 }
 
 
-def notebook_title(provider: str, project: str | None) -> str:
+def notebook_path(provider: str, project: str | None) -> tuple[str, str]:
-    """Derive a flat Joplin notebook title from provider and project name.
+    """Return (parent_notebook, child_notebook) for the given provider and project.
+
+    The parent is the top-level provider notebook; the child is the project name.
 
     Examples:
-        notebook_title("chatgpt", "no-project") → "ChatGPT - No Project"
+        notebook_path("chatgpt", None)             → ("AI-ChatGPT", "No Project")
-        notebook_title("claude", "budget-tracker") → "Claude - Budget Tracker"
+        notebook_path("chatgpt", "no-project")     → ("AI-ChatGPT", "No Project")
-        notebook_title("chatgpt", None) → "ChatGPT - No Project"
+        notebook_path("claude", "budget-tracker")  → ("AI-Claude", "Budget Tracker")
     """
-    prov_display = _PROVIDER_DISPLAY.get(provider, provider.capitalize())
+    parent = _PROVIDER_DISPLAY.get(provider, f"AI-{provider.capitalize()}")
-    proj = (project or "no-project").replace("-", " ").title()
+    child = (project or "no-project").replace("-", " ").title()
-    return f"{prov_display} - {proj}"
+    return parent, child

src/loss_report.py

@@ -0,0 +1,123 @@
+"""Per-export-run tally for content that was dropped or partially extracted.
+
+Surfaces the loss visibility that the rest of the system promises in its
+output (visible ``unknown`` blocks). The summary emitted at the end of
+each export is the load-bearing operator-facing signal: if a real content
+type starts being silently dropped, this is where it shows up.
+
+Pass a single instance through ``BaseProvider.normalize_conversation`` and
+read it back in ``src/main.py`` after the export loop. No global state.
+"""
+
+from collections import Counter
+from dataclasses import dataclass, field
+
+_TOP_N_BREAKDOWN = 5
+
+
+@dataclass
+class LossReport:
+    """Counters for things that didn't render cleanly in an export run."""
+
+    # Type-keyed counters. Values are int counts.
+    unknown_blocks: Counter = field(default_factory=Counter)
+    extraction_failures: Counter = field(default_factory=Counter)
+    filtered_roles: Counter = field(default_factory=Counter)
+    # Messages collapsed/omitted by the EXPORTER_HIDDEN_CONTENT policy —
+    # intentional, but still surfaced so the omission is never invisible.
+    collapsed: Counter = field(default_factory=Counter)
+    collapsed_bytes: int = 0
+
+    # Aggregate counters
+    messages_rendered: int = 0
+    conversations: int = 0
+
+    # Media downloads (EXPORTER_DOWNLOAD_MEDIA): successes / skips / failures
+    media_downloaded: int = 0
+    media_downloaded_bytes: int = 0
+    media_failed: Counter = field(default_factory=Counter)
+
+    # Recording -------------------------------------------------------------
+
+    def record_unknown(self, raw_type: str) -> None:
+        self.unknown_blocks[raw_type or "?"] += 1
+
+    def record_extraction_failure(self, raw_type: str) -> None:
+        self.extraction_failures[raw_type or "?"] += 1
+
+    def record_filtered_role(self, role: str) -> None:
+        self.filtered_roles[role or "?"] += 1
+
+    def record_collapsed(self, origin: str, size_bytes: int = 0) -> None:
+        self.collapsed[origin or "?"] += 1
+        if isinstance(size_bytes, int) and size_bytes > 0:
+            self.collapsed_bytes += size_bytes
+
+    def record_media_downloaded(self, size_bytes: int = 0) -> None:
+        self.media_downloaded += 1
+        if isinstance(size_bytes, int) and size_bytes > 0:
+            self.media_downloaded_bytes += size_bytes
+
+    def record_media_failed(self, reason: str) -> None:
+        self.media_failed[reason or "?"] += 1
+
+    def record_message(self) -> None:
+        self.messages_rendered += 1
+
+    def record_conversation(self) -> None:
+        self.conversations += 1
+
+    # Summary ---------------------------------------------------------------
+
+    def format_summary(self) -> str:
+        """Return a multi-line summary table suitable for INFO logging.
+
+        Format pinned by plan §Post-export summary — "(none)" sentinel when a
+        counter is empty, top-5 breakdown with "+ N more types" overflow.
+        """
+        lines: list[str] = ["[export] Run summary:"]
+        lines.append(f"  conversations:        {self.conversations}")
+        lines.append(f"  messages rendered:    {self.messages_rendered}")
+        lines.extend(_format_section("unknown blocks:      ", self.unknown_blocks))
+        lines.extend(_format_section("extraction failures: ", self.extraction_failures))
+        lines.extend(_format_section("collapsed by policy: ", self.collapsed))
+        if self.collapsed:
+            lines.append(
+                f"                        ≈{self.collapsed_bytes / 1024:.0f} KB omitted "
+                "(intentional — EXPORTER_HIDDEN_CONTENT=full to keep)"
+            )
+        if self.media_downloaded or self.media_failed:
+            lines.append(
+                f"  media downloaded:     {self.media_downloaded} "
+                f"(≈{self.media_downloaded_bytes / 1024:.0f} KB)"
+            )
+            if self.media_failed:
+                total_failed = sum(self.media_failed.values())
+                lines.append(f"  media failed:         {total_failed}")
+                for reason, count in self.media_failed.most_common(_TOP_N_BREAKDOWN):
+                    lines.append(f"                        {reason}={count}")
+        lines.append(
+            "  filtered roles:       "
+            "(filter lifted in v0.4.0 — counter retained for future use, expected 0)"
+        )
+        if self.filtered_roles:
+            for role, count in self.filtered_roles.most_common(_TOP_N_BREAKDOWN):
+                lines.append(f"                        {role}={count}")
+        return "\n".join(lines)
+
+
+def _format_section(label: str, counter: Counter) -> list[str]:
+    """Render one counter section: header line + indented breakdown lines."""
+    total = sum(counter.values())
+    header = f"  {label} {total}"
+    if total == 0:
+        return [header, "                        (none)"]
+
+    lines = [header]
+    most_common = counter.most_common()
+    for raw_type, count in most_common[:_TOP_N_BREAKDOWN]:
+        lines.append(f"                        {raw_type}={count}")
+    if len(most_common) > _TOP_N_BREAKDOWN:
+        remainder = len(most_common) - _TOP_N_BREAKDOWN
+        lines.append(f"                        + {remainder} more types")
+    return lines

src/main.py

@@ -2,11 +2,12 @@
 
 import importlib.metadata
 import logging
+import os
 import platform
 import shutil
 import sys
 import traceback
-from datetime import datetime
+from datetime import datetime, timezone
 from pathlib import Path
 
 import click
@@ -16,7 +17,8 @@ from rich.table import Table
 from src.cache import Cache, CacheError
 from src.config import ConfigError
 from src.logging_config import setup_logging
-from src.providers.base import ProviderError
+from src.loss_report import LossReport
+from src.providers.base import DRIFT_ERROR, DRIFT_OK, DRIFT_WARN, ProviderError
 
 console = Console()
 err_console = Console(stderr=True)
@@ -60,6 +62,14 @@ def cli(ctx: click.Context, verbose: bool, quiet: bool, debug: bool, no_log_file
     """Export ChatGPT and Claude conversations to Markdown for personal archival."""
     ctx.ensure_object(dict)
 
+    # Load .env BEFORE reading any env var (LOG_FILE, CACHE_DIR). Without this
+    # the cache/log paths fell back to defaults regardless of .env — the cache
+    # in particular silently lived at ./cache even when CACHE_DIR was set.
+    # override=False so a real shell env var still wins over .env.
+    import os
+    from dotenv import load_dotenv
+    load_dotenv(override=False)
+
     # Determine console log level
     if debug or verbose:
         level = logging.DEBUG
@@ -69,7 +79,6 @@ def cli(ctx: click.Context, verbose: bool, quiet: bool, debug: bool, no_log_file
         level = logging.INFO
 
     # Determine log file path from env (setup_logging handles "none")
-    import os
     log_file = os.getenv("LOG_FILE", "./cache/logs/exporter.log")
 
     setup_logging(level=level, log_file=log_file, no_log_file=no_log_file)
@@ -144,8 +153,6 @@ def auth(ctx: click.Context) -> None:
 
 
 def _auth_chatgpt(os_name: str) -> None:
-    import jwt as pyjwt
-
     console.print("\n[bold]─── ChatGPT ───[/bold]")
     console.print("1. Open [link=https://chatgpt.com]https://chatgpt.com[/link] and log in.")
     if os_name == "Darwin":
@@ -166,22 +173,12 @@ def _auth_chatgpt(os_name: str) -> None:
 
     token_1 = click.prompt("ChatGPT session token (.1, leave blank if absent)", hide_input=True, default="", show_default=False).strip() or None
 
-    # Validate
+    # Format hint only — the token is a JWE whose expiry can't be decoded
+    # client-side, so validity is judged by the live check below (the
+    # /api/auth/session `error` field), not by parsing the token. See §9.
     if not token.startswith("eyJ"):
         console.print("[yellow]Warning: token doesn't look like a JWT (expected 'eyJ...').[/yellow]")
 
-    expiry_str = ""
-    try:
-        payload = pyjwt.decode(token, options={"verify_signature": False})
-        exp = payload.get("exp")
-        if exp:
-            from datetime import timezone
-            expiry = datetime.fromtimestamp(exp, tz=timezone.utc)
-            expiry_str = expiry.strftime("%Y-%m-%d %H:%M UTC")
-            console.print(f"[green]Token decoded — expires: {expiry_str}[/green]")
-    except Exception:
-        console.print("[yellow]Could not decode token expiry.[/yellow]")
-
     # Live validation — exchange session token for an access token
     _valid = False
     _error: str | None = None
@@ -280,36 +277,39 @@ def _auth_claude(os_name: str) -> None:
 
 def _write_token_to_env(key: str, value: str) -> None:
     """Write or update a key in .env, offering to create the file if it doesn't exist."""
-    env_path = Path(".env")
-
     if click.confirm(f"Write {key} to .env?", default=True):
-        if not env_path.exists():
+        _set_env_key(key, value)
-            # Create from example if available
-            example = Path(".env.example")
-            if example.exists():
-                import shutil as _shutil
-                _shutil.copy2(example, env_path)
-                console.print("[dim]Created .env from .env.example[/dim]")
-            else:
-                env_path.touch()
 
-        lines = env_path.read_text(encoding="utf-8").splitlines(keepends=True)
-        updated = False
-        new_lines = []
-        for line in lines:
-            if line.startswith(f"{key}=") or line.startswith(f"{key} ="):
-                new_lines.append(f"{key}={value}\n")
-                updated = True
-            else:
-                new_lines.append(line)
 
-        if not updated:
+def _set_env_key(key: str, value: str) -> None:
-            new_lines.append(f"\n{key}={value}\n")
+    """Write or update a key in .env without prompting (creates .env if needed)."""
+    env_path = Path(".env")
+    if not env_path.exists():
+        # Create from example if available
+        example = Path(".env.example")
+        if example.exists():
+            import shutil as _shutil
+            _shutil.copy2(example, env_path)
+            console.print("[dim]Created .env from .env.example[/dim]")
+        else:
+            env_path.touch()
 
-        env_path.write_text("".join(new_lines), encoding="utf-8")
+    lines = env_path.read_text(encoding="utf-8").splitlines(keepends=True)
-        import os
+    updated = False
-        os.chmod(env_path, 0o600)
+    new_lines = []
-        console.print(f"[green]{key} written to .env (permissions: 600)[/green]")
+    for line in lines:
+        if line.startswith(f"{key}=") or line.startswith(f"{key} ="):
+            new_lines.append(f"{key}={value}\n")
+            updated = True
+        else:
+            new_lines.append(line)
+
+    if not updated:
+        new_lines.append(f"\n{key}={value}\n")
+
+    env_path.write_text("".join(new_lines), encoding="utf-8")
+    os.chmod(env_path, 0o600)
+    console.print(f"[green]{key} written to .env (permissions: 600)[/green]")
 
 
 # ──────────────────────────────────────────────────────────────────────────────
@@ -325,18 +325,105 @@ def doctor(ctx: click.Context) -> None:
     Checks token presence, format, expiry, directory permissions, disk space,
     and live API reachability. Exits with code 1 if any checks fail.
     """
-    checks = _run_doctor_checks()
+    checks = _run_doctor_checks(cache=ctx.obj["cache"])
     _print_doctor_table(checks)
 
     if any(not c["pass"] for c in checks):
         sys.exit(1)
 
 
-def _run_doctor_checks() -> list[dict]:
+@cli.command()
-    """Run all doctor checks and return results."""
+@click.option(
+    "--provider",
+    type=click.Choice(["chatgpt", "claude", "all"]),
+    default="all",
+    help="Which web-API provider(s) to probe.",
+)
+@click.pass_context
+def canary(ctx: click.Context, provider: str) -> None:
+    """Probe provider APIs for schema drift against the fields the parser needs.
+
+    Fetches one listing page + one conversation per provider and checks only
+    the normalizer's load-bearing fields (not the full response shape, which
+    churns harmlessly). Surfaces the silent-failure risks for a backup tool:
+    a renamed retrieval-tool author bypassing the hidden-content collapse, a
+    new content_type, drifted message fields, or ignored attachments.
+
+    Exits non-zero if any ERROR finding is reported (a load-bearing field is
+    missing/mistyped). WARN findings are surfaced but do not fail. See
+    FUTURE.md §10.
+    """
+    from dotenv import load_dotenv
+
+    load_dotenv(override=False)
+
+    chatgpt_token = os.getenv("CHATGPT_SESSION_TOKEN", "").strip() or None
+    claude_key = os.getenv("CLAUDE_SESSION_KEY", "").strip() or None
+
+    targets: list[tuple[str, object]] = []
+    if provider in ("chatgpt", "all") and chatgpt_token:
+        from src.providers.chatgpt import ChatGPTProvider
+
+        ct1 = os.getenv("CHATGPT_SESSION_TOKEN_1", "").strip() or None
+        targets.append(
+            ("chatgpt", lambda: ChatGPTProvider(chatgpt_token, session_token_1=ct1))
+        )
+    if provider in ("claude", "all") and claude_key:
+        from src.providers.claude import ClaudeProvider
+
+        targets.append(("claude", lambda: ClaudeProvider(claude_key)))
+
+    if not targets:
+        err_console.print(
+            "[yellow]No web-API provider tokens configured "
+            "(set CHATGPT_SESSION_TOKEN / CLAUDE_SESSION_KEY).[/yellow]"
+        )
+        sys.exit(1)
+
+    rows: list[tuple[str, dict]] = []
+    for name, factory in targets:
+        try:
+            findings = factory().check_drift()
+        except ProviderError as e:
+            findings = [{"severity": DRIFT_ERROR, "check": "probe", "detail": str(e.original)[:100]}]
+        except Exception as e:
+            findings = [{"severity": DRIFT_ERROR, "check": "probe", "detail": str(e)[:100]}]
+        for f in findings:
+            rows.append((name, f))
+
+    _print_canary_table(rows)
+    if any(f["severity"] == DRIFT_ERROR for _, f in rows):
+        sys.exit(1)
+
+
+def _print_canary_table(rows: list[tuple[str, dict]]) -> None:
+    table = Table(title="API Drift Canary", show_header=True)
+    table.add_column("Provider", style="bold")
+    table.add_column("Severity", justify="center")
+    table.add_column("Check")
+    table.add_column("Detail")
+    badge = {
+        DRIFT_OK: "[green]OK[/green]",
+        DRIFT_WARN: "[yellow]WARN[/yellow]",
+        DRIFT_ERROR: "[red]ERROR[/red]",
+    }
+    for name, f in rows:
+        table.add_row(
+            name,
+            badge.get(f["severity"], f["severity"]),
+            f.get("check", ""),
+            f.get("detail", ""),
+        )
+    console.print(table)
+
+
+def _run_doctor_checks(cache: Cache | None = None) -> list[dict]:
+    """Run all doctor checks and return results.
+
+    ``cache``: when provided, also verify manifest ↔ disk integrity (every
+    ``file_path`` recorded in the manifest exists on disk).
+    """
     import os
-    import jwt as pyjwt
-    from datetime import timezone
     from dotenv import load_dotenv
 
     # Load .env so doctor works without the user having to export vars manually
@@ -358,25 +445,21 @@ def _run_doctor_checks() -> list[dict]:
     if chatgpt_token:
         is_jwt = chatgpt_token.startswith("eyJ")
         add("ChatGPT token is valid JWT", is_jwt, "" if is_jwt else "Expected token starting with 'eyJ'")
+        # Expiry is NOT decodable from the token: ChatGPT's is a JWE (encrypted),
+        # so the `exp` claim is sealed server-side. The honest, free validity
+        # signal is the `error` field of /api/auth/session — `RefreshAccessTokenError`
+        # means the session token is dead while `expires`/`accessToken` still
+        # echo stale values. See FUTURE.md §9.
         if is_jwt:
+            from src.providers.chatgpt import ChatGPTProvider
+            ct1 = os.getenv("CHATGPT_SESSION_TOKEN_1", "").strip() or None
             try:
-                payload = pyjwt.decode(chatgpt_token, options={"verify_signature": False})
+                ok, detail = ChatGPTProvider(
-                exp = payload.get("exp")
+                    chatgpt_token, session_token_1=ct1
-                if exp:
+                ).session_health()
-                    expiry = datetime.fromtimestamp(exp, tz=timezone.utc)
+                add("ChatGPT token active", ok, detail)
-                    now = datetime.now(tz=timezone.utc)
+            except Exception as e:
-                    delta = expiry - now
+                add("ChatGPT token active", False, str(e)[:80])
-                    detail = f"Expires {expiry.strftime('%Y-%m-%d %H:%M UTC')} ({delta.days}d)"
-                    ok = delta.total_seconds() > 0
-                    add("ChatGPT token not expired", ok, detail)
-                    if ok and delta.total_seconds() < 86400:
-                        add("ChatGPT token expiry warning", False, "Expires in < 24h — refresh soon")
-                else:
-                    add("ChatGPT token expiry", False, "JWT has no 'exp' claim")
-            except pyjwt.exceptions.DecodeError:
-                # JWE (encrypted JWT) — cannot decode without the server key.
-                # This is normal for ChatGPT's current token format. Token is present and valid.
-                add("ChatGPT token expiry", True, "Encrypted token (JWE) — expiry not decodable client-side")
 
     # Claude key
     if claude_key:
@@ -404,6 +487,18 @@ def _run_doctor_checks() -> list[dict]:
     except OSError as e:
         add("Disk space check", False, str(e))
 
+    # Manifest ↔ disk integrity
+    if cache is not None:
+        referenced = cache.all_file_paths()
+        if referenced:
+            missing = sorted(
+                p for p in referenced if not Path(p).expanduser().exists()
+            )
+            detail = f"{len(referenced) - len(missing)}/{len(referenced)} manifest files present"
+            if missing:
+                detail += f"; first missing: {missing[0]} — re-export to fix"
+            add("Manifest files on disk", not missing, detail)
+
     # API reachability
     if chatgpt_token:
         try:
@@ -452,7 +547,7 @@ def _print_doctor_table(checks: list[dict]) -> None:
 @cli.command()
 @click.option(
     "--provider",
-    type=click.Choice(["chatgpt", "claude", "all"], case_sensitive=False),
+    type=click.Choice(["chatgpt", "claude", "claude-code", "all"], case_sensitive=False),
     default="all",
     show_default=True,
     help="Which provider to export.",
@@ -486,6 +581,47 @@ def _print_doctor_table(checks: list[dict]) -> None:
         "Use 'none' for conversations outside any project."
     ),
 )
+@click.option(
+    "--hidden-content",
+    type=click.Choice(["full", "placeholder", "omit"], case_sensitive=False),
+    default=None,
+    help=(
+        "What to do with content that was invisible in the provider's web UI "
+        "(file-retrieval tool dumps, hidden context): keep in full, collapse to "
+        "a one-line placeholder, or omit. Overrides EXPORTER_HIDDEN_CONTENT "
+        "(default: placeholder)."
+    ),
+)
+@click.option(
+    "--max-conversations",
+    type=click.IntRange(min=1),
+    default=None,
+    help=(
+        "Cap how many conversations are downloaded this run (per provider). "
+        "Runs are resumable, so re-running continues where the cap stopped. "
+        "Overrides MAX_CONVERSATIONS_PER_RUN (default: unlimited)."
+    ),
+)
+@click.option(
+    "--download-media",
+    type=click.Choice(["images", "all", "off"], case_sensitive=False),
+    default=None,
+    help=(
+        "Download conversation assets next to the exports: images only, "
+        "all (also audio/files), or off. Overrides EXPORTER_DOWNLOAD_MEDIA "
+        "(default: images)."
+    ),
+)
+@click.option(
+    "--force",
+    is_flag=True,
+    help=(
+        "Re-export every conversation even if already cached and unchanged. "
+        "Use to re-render the whole archive after a formatting/feature change "
+        "without 'cache --clear' (which would drop Joplin note links). Combine "
+        "with --max-conversations to spread the work over several runs."
+    ),
+)
 @click.option("--dry-run", is_flag=True, help="Show what would be exported without writing anything.")
 @click.pass_context
 def export(
@@ -495,6 +631,10 @@ def export(
     output_dir: str | None,
     since: str | None,
     project_filter: str | None,
+    hidden_content: str | None,
+    max_conversations: int | None,
+    download_media: str | None,
+    force: bool,
     dry_run: bool,
 ) -> None:
     """Export new and updated conversations to Markdown or JSON.
@@ -506,6 +646,13 @@ def export(
     debug = ctx.obj.get("debug", False)
     cache: Cache = ctx.obj["cache"]
 
+    # CLI flags win over .env: set before load_config (load_dotenv uses
+    # override=False, so an existing env var is not clobbered by .env).
+    if hidden_content:
+        os.environ["EXPORTER_HIDDEN_CONTENT"] = hidden_content.lower()
+    if download_media:
+        os.environ["EXPORTER_DOWNLOAD_MEDIA"] = download_media.lower()
+
     # Load config (may raise ConfigError)
     try:
         from src.config import load_config
@@ -516,7 +663,7 @@ def export(
     # First-run: auto-doctor
     if not cache.last_run():
         console.print("[dim]First run — checking configuration…[/dim]")
-        checks = _run_doctor_checks()
+        checks = _run_doctor_checks(cache=cache)
         _print_doctor_table(checks)
         if any(not c["pass"] for c in checks):
             err_console.print(
@@ -536,6 +683,21 @@ def export(
             err_console.print(f"[red]Invalid --since date: '{since}'. Use YYYY-MM-DD.[/red]")
             sys.exit(1)
 
+    # Session cap: CLI flag wins over MAX_CONVERSATIONS_PER_RUN env default.
+    session_cap = max_conversations if max_conversations is not None else cfg.max_conversations
+
+    # Force re-render campaign: stamp a start time so progress can be tracked
+    # across runs. A conversation is "done this campaign" once its exported_at
+    # passes the stamp; finished providers then export nothing, and the
+    # "still awaiting" count shrinks to zero. A fresh campaign starts whenever
+    # none is active (the previous one completed or was never started).
+    campaign_at: str | None = None
+    if force and not dry_run:
+        campaign_at = cache.get_force_campaign()
+        if campaign_at is None:
+            campaign_at = datetime.now(timezone.utc).isoformat()
+            cache.set_force_campaign(campaign_at)
+
     # Determine which providers to run
     providers_to_run = _resolve_providers(provider, cfg)
     if not providers_to_run:
@@ -554,6 +716,9 @@ def export(
     # Summary counters
     summary: dict[str, dict[str, int]] = {}
 
+    # Single LossReport tracks data-loss visibility across all providers in this run.
+    loss_report = LossReport()
+
     for prov_name, prov_instance in providers_to_run:
         summary[prov_name] = {"exported": 0, "skipped": 0, "failed": 0}
 
@@ -572,19 +737,44 @@ def export(
                 f"  [dim]--project filter '{project_filter}': {len(all_convs)} matching conversations.[/dim]"
             )
 
-        to_export = cache.get_new_or_updated(prov_name, all_convs)
+        to_export = cache.get_new_or_updated(
+            prov_name, all_convs, force=force, campaign_at=campaign_at
+        )
         skipped = len(all_convs) - len(to_export)
         summary[prov_name]["skipped"] = skipped
 
+        # Apply the session cap. Resumability makes this safe: the manifest
+        # records each conversation immediately, so the next run picks up
+        # exactly where the cap stopped.
+        deferred = 0
+        if session_cap is not None and len(to_export) > session_cap:
+            deferred = len(to_export) - session_cap
+            to_export = to_export[:session_cap]
+        summary[prov_name]["deferred"] = deferred
+
         if dry_run:
             _print_dry_run_table(prov_name, to_export, prov_instance, export_base, structure, skipped)
+            if deferred:
+                console.print(
+                    f"  [yellow]Session cap {session_cap}: {deferred} more pending beyond this run.[/yellow]"
+                )
             continue
 
         if not to_export:
-            console.print(f"  [dim]{skipped} conversations already up to date.[/dim]")
+            done_msg = (
+                "already re-rendered this campaign."
+                if force else "conversations already up to date."
+            )
+            console.print(f"  [dim]{skipped} {done_msg}[/dim]")
             continue
 
-        console.print(f"  [dim]{len(to_export)} to export, {skipped} already up to date.[/dim]")
+        if deferred:
+            console.print(
+                f"  [dim]{len(to_export)} to export (session cap; {deferred} deferred), "
+                f"{skipped} already up to date.[/dim]"
+            )
+        else:
+            console.print(f"  [dim]{len(to_export)} to export, {skipped} already up to date.[/dim]")
 
         from rich.progress import Progress, SpinnerColumn, TextColumn, BarColumn, TaskProgressColumn
 
@@ -611,7 +801,20 @@ def export(
                     for key, val in raw_conv.items():
                         if (key.startswith("_") or key in _PROPAGATE_KEYS) and key not in full_raw:
                             full_raw[key] = val
-                    normalized = prov_instance.normalize_conversation(full_raw)
+                    normalized = prov_instance.normalize_conversation(full_raw, loss_report)
+
+                    # Download assets (images by default) so the renderers
+                    # can inline them. Failures keep placeholders; never fatal.
+                    if cfg.download_media != "off":
+                        from src.media import resolve_media
+                        resolve_media(
+                            normalized,
+                            prov_instance,
+                            export_base,
+                            structure,
+                            cfg.download_media,
+                            loss_report,
+                        )
 
                     exported_path: Path | None = None
                     if md_exporter:
@@ -623,6 +826,7 @@ def export(
                     cache.mark_exported(prov_name, conv_id, {
                         "title": normalized.get("title", ""),
                         "project": normalized.get("project"),
+                        "created_at": normalized.get("created_at", ""),
                         "updated_at": normalized.get("updated_at", ""),
                         "file_path": str(exported_path) if exported_path else "",
                     })
@@ -640,8 +844,42 @@ def export(
                     progress.advance(task)
                     continue
 
+        # Force campaign: how many of this provider's conversations still
+        # predate the campaign stamp (i.e. not yet re-rendered this campaign).
+        if force and campaign_at is not None:
+            summary[prov_name]["awaiting"] = sum(
+                1
+                for c in all_convs
+                if cache.exported_at(prov_name, c.get("id") or c.get("uuid", "")) < campaign_at
+            )
+
     if not dry_run:
         _print_export_summary(summary)
+        if force:
+            total_awaiting = sum(s.get("awaiting", 0) for s in summary.values())
+            exported_now = sum(s.get("exported", 0) for s in summary.values())
+            if total_awaiting:
+                console.print(
+                    f"[yellow]Force re-render: refreshed {exported_now} this run, "
+                    f"{total_awaiting} still to go. Re-run the same command to continue.[/yellow]"
+                )
+            else:
+                cache.clear_force_campaign()
+                console.print(
+                    "[green]Force re-render complete — every conversation has been "
+                    "re-rendered with the current code.[/green]"
+                )
+        else:
+            total_deferred = sum(s.get("deferred", 0) for s in summary.values())
+            if total_deferred:
+                console.print(
+                    f"[yellow]{total_deferred} conversation(s) deferred by the session cap "
+                    f"({session_cap} per provider per run). Re-run the same command to continue.[/yellow]"
+                )
+        # Emit the data-loss summary at INFO level so it lands in the log file
+        # AND the operator's console (default level is INFO).
+        for line in loss_report.format_summary().split("\n"):
+            logger.info(line)
 
 
 def _resolve_providers(provider: str, cfg) -> list[tuple[str, object]]:
@@ -674,6 +912,7 @@ def _resolve_providers(provider: str, cfg) -> list[tuple[str, object]]:
                         session_token=cfg.chatgpt_session_token,
                         session_token_1=cfg.chatgpt_session_token_1,
                         project_ids=cfg.chatgpt_project_ids,
+                        hidden_content=cfg.hidden_content,
                     ),
                 ))
             except ProviderError as e:
@@ -685,6 +924,19 @@ def _resolve_providers(provider: str, cfg) -> list[tuple[str, object]]:
     if provider in ("claude", "all"):
         try_add("claude", cfg.claude_session_key, ClaudeProvider)
 
+    if provider in ("claude-code", "all"):
+        from src.providers.claude_code import ClaudeCodeProvider, DEFAULT_PROJECTS_DIR
+        cc_dir = Path(os.getenv("CLAUDE_CODE_DIR", DEFAULT_PROJECTS_DIR)).expanduser()
+        if cc_dir.is_dir():
+            result.append((
+                "claude-code",
+                ClaudeCodeProvider(projects_dir=cc_dir, hidden_content=cfg.hidden_content),
+            ))
+        elif provider == "claude-code":
+            logging.getLogger(__name__).warning(
+                "[claude-code] Skipping — %s not found (set CLAUDE_CODE_DIR).", cc_dir
+            )
+
     return result
 
 
@@ -781,7 +1033,7 @@ def _print_export_summary(summary: dict[str, dict[str, int]]) -> None:
 @cli.command(name="list")
 @click.option(
     "--provider",
-    type=click.Choice(["chatgpt", "claude", "all"], case_sensitive=False),
+    type=click.Choice(["chatgpt", "claude", "claude-code", "all"], case_sensitive=False),
     default="all",
     show_default=True,
 )
@@ -847,7 +1099,7 @@ def list_conversations(ctx: click.Context, provider: str, project_filter: str |
 @click.option("--clear", is_flag=True, help="Clear cached entries.")
 @click.option(
     "--provider",
-    type=click.Choice(["chatgpt", "claude", "all"], case_sensitive=False),
+    type=click.Choice(["chatgpt", "claude", "claude-code", "all"], case_sensitive=False),
     default="all",
     help="Provider to target (used with --clear).",
 )
@@ -877,6 +1129,106 @@ def cache(ctx: click.Context, show: bool, clear: bool, provider: str) -> None:
         console.print("Specify --show or --clear. Use --help for options.")
 
 
+# ──────────────────────────────────────────────────────────────────────────────
+# prune command
+# ──────────────────────────────────────────────────────────────────────────────
+
+
+@cli.command()
+@click.option("--dry-run", is_flag=True, help="List stale files without deleting anything.")
+@click.option("--yes", "-y", is_flag=True, help="Delete without a confirmation prompt.")
+@click.pass_context
+def prune(ctx: click.Context, dry_run: bool, yes: bool) -> None:
+    """Delete export files no longer referenced by the cache manifest.
+
+    Removes leftovers from old folder layouts and orphaned files from fixed
+    bugs (e.g. pre-v0.5.0 nested year folders, files with a missing
+    conversation ID). Only .md and .json files under EXPORT_DIR are
+    considered; the manifest is the source of truth.
+    """
+    cache_obj: Cache = ctx.obj["cache"]
+
+    from dotenv import load_dotenv
+    load_dotenv(override=False)
+    export_dir = Path(os.getenv("EXPORT_DIR", "./exports")).expanduser()
+
+    if not export_dir.exists():
+        console.print(f"[yellow]Export directory {export_dir} does not exist — nothing to prune.[/yellow]")
+        return
+
+    referenced_raw = cache_obj.all_file_paths()
+    if not referenced_raw:
+        # Footgun guard: with an empty manifest (e.g. right after `cache
+        # --clear`) every export file would look stale and prune would wipe
+        # the entire archive.
+        err_console.print(
+            "[red]The manifest references no export files (did you just run "
+            "'cache --clear'?). Refusing to prune — run an export first.[/red]"
+        )
+        sys.exit(1)
+
+    referenced = {Path(p).expanduser().resolve() for p in referenced_raw}
+    on_disk = sorted(
+        f for f in export_dir.rglob("*")
+        if f.is_file() and f.suffix in (".md", ".json")
+    )
+    stale = [f for f in on_disk if f.resolve() not in referenced]
+
+    if not stale:
+        console.print(
+            f"[green]Nothing to prune — all {len(on_disk)} export files are referenced "
+            "by the manifest.[/green]"
+        )
+        return
+
+    total_bytes = sum(f.stat().st_size for f in stale)
+    console.print(
+        f"[bold]{len(stale)} stale file(s)[/bold] "
+        f"({total_bytes / (1024 * 1024):.1f} MB) not referenced by the manifest "
+        f"(of {len(on_disk)} total):\n"
+    )
+    for f in stale:
+        try:
+            shown = f.relative_to(export_dir.resolve()) if f.is_absolute() else f
+        except ValueError:
+            shown = f
+        console.print(f"  [dim]{shown}[/dim]")
+
+    if dry_run:
+        console.print("\n[yellow]Dry run — nothing deleted.[/yellow]")
+        return
+
+    if not yes and not click.confirm(f"\nDelete these {len(stale)} files?", default=False):
+        console.print("Aborted — nothing deleted.")
+        return
+
+    deleted = 0
+    for f in stale:
+        try:
+            f.unlink()
+            deleted += 1
+        except OSError as e:
+            logger.error("[prune] Could not delete %s: %s", f, e)
+
+    # Sweep now-empty directories (deepest first).
+    removed_dirs = 0
+    for d in sorted(
+        (d for d in export_dir.rglob("*") if d.is_dir()),
+        key=lambda p: len(p.parts),
+        reverse=True,
+    ):
+        try:
+            d.rmdir()  # only succeeds when empty
+            removed_dirs += 1
+        except OSError:
+            pass
+
+    console.print(
+        f"[green]Deleted {deleted} file(s) ({total_bytes / (1024 * 1024):.1f} MB) "
+        f"and {removed_dirs} empty director(ies).[/green]"
+    )
+
+
 # ──────────────────────────────────────────────────────────────────────────────
 # joplin command
 # ──────────────────────────────────────────────────────────────────────────────
@@ -885,7 +1237,7 @@ def cache(ctx: click.Context, show: bool, clear: bool, provider: str) -> None:
 @cli.command()
 @click.option(
     "--provider",
-    type=click.Choice(["chatgpt", "claude", "all"], case_sensitive=False),
+    type=click.Choice(["chatgpt", "claude", "claude-code", "all"], case_sensitive=False),
     default="all",
     show_default=True,
     help="Which provider's conversations to sync to Joplin.",
@@ -908,9 +1260,9 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
     via its local REST API. Requires Joplin desktop to be running with the
     Web Clipper service enabled.
 
-    Notebooks are created automatically based on provider and project:
+    Notebooks are created automatically as nested folders:
-      exports/chatgpt/my-project/  →  "ChatGPT - My Project" notebook
+      chatgpt / my-project  →  AI-ChatGPT / My Project
-      exports/claude/no-project/   →  "Claude - No Project" notebook
+      claude  / no-project  →  AI-Claude   / No Project
 
     Re-running is safe: notes are updated (not duplicated) on subsequent runs.
 
@@ -936,7 +1288,7 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
         )
         sys.exit(1)
 
-    from src.joplin import JoplinClient, JoplinError, notebook_title
+    from src.joplin import JoplinClient, JoplinError, notebook_path
 
     client = JoplinClient(cfg.joplin_api_url, cfg.joplin_api_token)
 
@@ -958,10 +1310,9 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
 
     # Determine which providers to process
     providers_to_sync: list[str] = []
-    if provider in ("chatgpt", "all"):
+    for prov in ("chatgpt", "claude", "claude-code"):
-        providers_to_sync.append("chatgpt")
+        if provider in (prov, "all"):
-    if provider in ("claude", "all"):
+            providers_to_sync.append(prov)
-        providers_to_sync.append("claude")
 
     summary: dict[str, dict[str, int]] = {}
 
@@ -1016,7 +1367,9 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
 
             for conv_id, entry in pending:
                 file_path = entry.get("file_path", "")
-                title = entry.get("title") or "Untitled"
+                raw_title = entry.get("title") or "Untitled"
+                created_date = (entry.get("created_at") or "")[:10]
+                title = f"{created_date} {raw_title}" if created_date else raw_title
                 project = entry.get("project") or None
                 existing_note_id = entry.get("joplin_note_id")
                 action = "update" if existing_note_id else "create"
@@ -1031,9 +1384,20 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
                     body = Path(file_path).read_text(encoding="utf-8")
                     logger.debug("[joplin] Read %d chars from %s", len(body), file_path)
 
-                    # Get or create the notebook
+                    # Upload any downloaded media as Joplin resources and
-                    nb_title = notebook_title(prov_name, project)
+                    # rewrite local media/ links to :/resourceId references.
-                    notebook_id = client.get_or_create_notebook(nb_title)
+                    # The cache stores the resource map so re-syncs reuse IDs.
+                    from src.joplin import upload_media_and_rewrite
+                    res_map = cache_obj.get_joplin_resources(prov_name, conv_id)
+                    body, new_res_map = upload_media_and_rewrite(
+                        body, Path(file_path).parent, client, res_map
+                    )
+                    if new_res_map != res_map:
+                        cache_obj.set_joplin_resources(prov_name, conv_id, new_res_map)
+
+                    # Get or create the nested notebook
+                    nb_path = notebook_path(prov_name, project)
+                    notebook_id = client.get_or_create_notebook_path(list(nb_path))
 
                     if existing_note_id:
                         client.update_note(existing_note_id, title, body)
@@ -1070,7 +1434,7 @@ def joplin(ctx: click.Context, provider: str, project_filter: str | None, dry_ru
 
 
 def _print_joplin_dry_run_table(prov_name: str, pending: list[tuple[str, dict]]) -> None:
-    from src.joplin import notebook_title
+    from src.joplin import notebook_path
 
     table = Table(title=f"[DRY RUN] {prov_name.upper()} — Would sync {len(pending)} conversation(s)")
     table.add_column("Title")
@@ -1079,9 +1443,12 @@ def _print_joplin_dry_run_table(prov_name: str, pending: list[tuple[str, dict]])
     table.add_column("Action")
 
     for conv_id, entry in pending[:50]:
-        title = entry.get("title") or "Untitled"
+        raw_title = entry.get("title") or "Untitled"
+        created_date = (entry.get("created_at") or "")[:10]
+        title = f"{created_date} {raw_title}" if created_date else raw_title
         project = entry.get("project") or "no-project"
-        nb = notebook_title(prov_name, entry.get("project"))
+        parent, child = notebook_path(prov_name, entry.get("project"))
+        nb = f"{parent} / {child}"
         action = "update" if entry.get("joplin_note_id") else "create"
         table.add_row(title[:50], project[:30], nb, action)
 

src/media.py

@@ -0,0 +1,202 @@
+"""Media resolution — download conversation assets next to the export files.
+
+Walks a normalized conversation's placeholder blocks, downloads each asset via
+the provider, saves it under a ``media/`` directory beside the conversation's
+Markdown file, and annotates the block with a relative ``local_path`` so the
+renderer emits a real inline image / file link instead of a placeholder.
+
+Policy (``EXPORTER_DOWNLOAD_MEDIA``):
+  images (default) — image_placeholder blocks only
+  all              — also file_placeholder blocks (voice-mode audio etc.;
+                     measured 2026-06-12: 556 clips ≈ 162MB whose transcripts
+                     are already in the exports as text)
+  off              — leave placeholders untouched
+
+Failures (expired assets, network) keep the existing placeholder and are
+counted in the run summary — never fatal to the export.
+"""
+
+import logging
+import os
+import tempfile
+from pathlib import Path
+
+from src.blocks import BLOCK_TYPE_FILE_PLACEHOLDER, BLOCK_TYPE_IMAGE_PLACEHOLDER
+from src.loss_report import LossReport
+from src.providers.base import ProviderError
+from src.utils import build_export_path, generate_filename
+
+logger = logging.getLogger(__name__)
+
+MEDIA_IMAGES = "images"
+MEDIA_ALL = "all"
+MEDIA_OFF = "off"
+VALID_MEDIA_POLICIES = {MEDIA_IMAGES, MEDIA_ALL, MEDIA_OFF}
+
+_EXT_BY_MIME = {
+    "image/png": ".png",
+    "image/jpeg": ".jpg",
+    "image/gif": ".gif",
+    "image/webp": ".webp",
+    "audio/wav": ".wav",
+    "audio/x-wav": ".wav",
+    "audio/mpeg": ".mp3",
+    "audio/mp4": ".m4a",
+    "audio/aac": ".aac",
+    "application/pdf": ".pdf",
+}
+
+
+def resolve_media_policy() -> str:
+    """Read EXPORTER_DOWNLOAD_MEDIA from the environment, defaulting to images."""
+    value = os.getenv("EXPORTER_DOWNLOAD_MEDIA", "").strip().lower()
+    if not value:
+        return MEDIA_IMAGES
+    if value not in VALID_MEDIA_POLICIES:
+        logger.warning(
+            "EXPORTER_DOWNLOAD_MEDIA=%r is invalid (expected images|all|off) "
+            "— using 'images'.",
+            value,
+        )
+        return MEDIA_IMAGES
+    return value
+
+
+def resolve_media(
+    normalized: dict,
+    provider,
+    export_base: Path,
+    structure: str,
+    policy: str,
+    report: LossReport,
+) -> int:
+    """Download assets for one normalized conversation. Returns download count.
+
+    Mutates placeholder blocks in place (adds ``local_path``). Idempotent:
+    an asset already on disk is annotated without an API call.
+    """
+    if policy == MEDIA_OFF:
+        return 0
+
+    download = getattr(provider, "download_asset", None)
+    if download is None:
+        # Provider has no remote assets (e.g. claude-code local transcripts).
+        return 0
+
+    wanted_types = {BLOCK_TYPE_IMAGE_PLACEHOLDER}
+    if policy == MEDIA_ALL:
+        wanted_types.add(BLOCK_TYPE_FILE_PLACEHOLDER)
+
+    targets = [
+        block
+        for message in normalized.get("messages", [])
+        for block in message.get("blocks", [])
+        if block.get("type") in wanted_types and block.get("ref")
+    ]
+    if not targets:
+        return 0
+
+    # Same path computation the exporter uses, so media/ lands beside the .md.
+    filename = generate_filename(
+        normalized.get("title", "Untitled"),
+        normalized.get("id", ""),
+        normalized.get("created_at") or "2000-01-01",
+    )
+    conv_dir = build_export_path(
+        export_base,
+        normalized.get("provider", ""),
+        normalized.get("project"),
+        normalized.get("created_at") or "2000-01-01",
+        filename,
+        structure,
+    ).parent
+    media_dir = conv_dir / "media"
+
+    downloaded = 0
+    for block in targets:
+        ref = block["ref"]
+        file_id = _safe_asset_name(provider, ref)
+        if not file_id:
+            report.record_media_failed("unparseable-ref")
+            continue
+
+        existing = _find_existing(media_dir, file_id)
+        if existing is not None:
+            block["local_path"] = f"media/{existing.name}"
+            continue
+
+        try:
+            content, mime, file_name = download(ref)
+        except ProviderError as e:
+            logger.warning(
+                "[media] Could not download %s: %s", ref[:60], e.original
+            )
+            reason = "expired-or-missing" if "404" in str(e.original) or "not found" in str(
+                e.original
+            ).lower() else "download-error"
+            report.record_media_failed(reason)
+            continue
+
+        ext = _pick_extension(mime, file_name)
+        target = media_dir / f"{file_id}{ext}"
+        try:
+            _write_atomic(target, content)
+        except OSError as e:
+            logger.error("[media] Could not write %s: %s", target, e)
+            report.record_media_failed("write-error")
+            continue
+
+        block["local_path"] = f"media/{target.name}"
+        report.record_media_downloaded(len(content))
+        downloaded += 1
+
+    return downloaded
+
+
+def _safe_asset_name(provider, ref: str) -> str | None:
+    """A stable, filesystem-safe name for the asset (the provider file ID)."""
+    parser = getattr(provider, "parse_asset_file_id", None)
+    if parser is None:
+        from src.providers.chatgpt import parse_asset_file_id as parser
+    file_id = parser(ref)
+    if not file_id:
+        return None
+    return "".join(c if c.isalnum() or c in "-_." else "_" for c in file_id)
+
+
+def _find_existing(media_dir: Path, file_id: str) -> Path | None:
+    """Return an already-downloaded file for this asset, if present."""
+    if not media_dir.is_dir():
+        return None
+    for candidate in media_dir.glob(f"{file_id}.*"):
+        if candidate.is_file() and candidate.stat().st_size > 0:
+            return candidate
+    return None
+
+
+def _pick_extension(mime: str | None, file_name: str | None) -> str:
+    if mime:
+        base_mime = mime.split(";")[0].strip().lower()
+        if base_mime in _EXT_BY_MIME:
+            return _EXT_BY_MIME[base_mime]
+    if file_name:
+        suffix = Path(file_name).suffix
+        if suffix and len(suffix) <= 8:
+            return suffix.lower()
+    return ".bin"
+
+
+def _write_atomic(target: Path, content: bytes) -> None:
+    target.parent.mkdir(parents=True, exist_ok=True)
+    fd, tmp_name = tempfile.mkstemp(dir=target.parent, suffix=".tmp")
+    try:
+        with os.fdopen(fd, "wb") as fh:
+            fh.write(content)
+        os.chmod(tmp_name, 0o600)
+        os.replace(tmp_name, target)
+    except OSError:
+        try:
+            os.unlink(tmp_name)
+        except OSError:
+            pass
+        raise

src/providers/base.py

@@ -1,6 +1,7 @@
 """Abstract base class for AI chat providers."""
 
 import logging
+import os
 import random
 import time
 from abc import ABC, abstractmethod
@@ -9,6 +10,7 @@ from typing import Any
 
 import requests
 
+from src.loss_report import LossReport
 from src.utils import redact_secrets
 
 # curl_cffi has its own exception hierarchy (rooted at CurlError → OSError),
@@ -36,6 +38,83 @@ MAX_RETRIES = 3
 BACKOFF_BASE = 2.0
 BACKOFF_MAX = 60.0
 
+# EXPORTER_HIDDEN_CONTENT policy — what to do with content that was invisible
+# in the provider's UI (retrieval dumps, hidden-flagged context, agent tool
+# traffic). Shared by all providers.
+HIDDEN_CONTENT_FULL = "full"
+HIDDEN_CONTENT_PLACEHOLDER = "placeholder"
+HIDDEN_CONTENT_OMIT = "omit"
+VALID_HIDDEN_CONTENT_POLICIES = {
+    HIDDEN_CONTENT_FULL,
+    HIDDEN_CONTENT_PLACEHOLDER,
+    HIDDEN_CONTENT_OMIT,
+}
+
+# ---------------------------------------------------------------------------
+# API-drift canary (FUTURE.md §10)
+# ---------------------------------------------------------------------------
+# The export depends on undocumented internal web APIs that can change shape
+# without notice. The canary fetches a small live sample and asserts only the
+# normalizer's load-bearing fields (NOT full shape — provider settings objects
+# churn constantly and would be pure noise). Findings carry a severity:
+#   ERROR — a load-bearing field is missing/mistyped; the parser will break or
+#           silently lose data. The backup is no longer trustworthy.
+#   WARN  — something unfamiliar appeared (new content_type, new tool author,
+#           a dormant field went live). Investigate; not necessarily broken.
+#   OK    — the assertion held.
+DRIFT_ERROR = "error"
+DRIFT_WARN = "warn"
+DRIFT_OK = "ok"
+
+
+def drift_finding(severity: str, check: str, detail: str = "") -> dict:
+    """Build one canary finding."""
+    return {"severity": severity, "check": check, "detail": detail}
+
+
+def resolve_hidden_content_policy() -> str:
+    """Read EXPORTER_HIDDEN_CONTENT from the environment, defaulting to placeholder."""
+    value = os.getenv("EXPORTER_HIDDEN_CONTENT", "").strip().lower()
+    if not value:
+        return HIDDEN_CONTENT_PLACEHOLDER
+    if value not in VALID_HIDDEN_CONTENT_POLICIES:
+        logger.warning(
+            "EXPORTER_HIDDEN_CONTENT=%r is invalid (expected full|placeholder|omit) "
+            "— using 'placeholder'.",
+            value,
+        )
+        return HIDDEN_CONTENT_PLACEHOLDER
+    return value
+
+
+# Default polite pacing between consecutive API requests (seconds). Keeps
+# traffic human-paced instead of bursty. REQUEST_DELAY=0 disables.
+DEFAULT_REQUEST_DELAY = 1.0
+
+
+def resolve_request_delay() -> float:
+    """Read REQUEST_DELAY from the environment, defaulting to DEFAULT_REQUEST_DELAY."""
+    raw = os.getenv("REQUEST_DELAY", "").strip()
+    if not raw:
+        return DEFAULT_REQUEST_DELAY
+    try:
+        value = float(raw)
+    except ValueError:
+        logger.warning(
+            "REQUEST_DELAY=%r is not a number — using default %.1fs.",
+            raw,
+            DEFAULT_REQUEST_DELAY,
+        )
+        return DEFAULT_REQUEST_DELAY
+    if value < 0:
+        logger.warning(
+            "REQUEST_DELAY=%r is negative — using default %.1fs.",
+            raw,
+            DEFAULT_REQUEST_DELAY,
+        )
+        return DEFAULT_REQUEST_DELAY
+    return value
+
 # Realistic Chrome User-Agent
 USER_AGENT = (
     "Mozilla/5.0 (X11; Linux x86_64) "
@@ -75,6 +154,9 @@ class BaseProvider(ABC):
                 "Accept-Language": "en-US,en;q=0.9",
             }
         )
+        # Polite pacing between consecutive requests (REQUEST_DELAY env).
+        self._request_delay = resolve_request_delay()
+        self._last_request_at: float | None = None
 
     # ------------------------------------------------------------------
     # Abstract interface — subclasses must implement these
@@ -89,13 +171,45 @@ class BaseProvider(ABC):
         """Return the full conversation detail for a single ID."""
 
     @abstractmethod
-    def normalize_conversation(self, raw: dict) -> dict:
+    def normalize_conversation(self, raw: dict, loss_report: LossReport | None = None) -> dict:
-        """Transform provider-specific schema to the common normalized schema."""
+        """Transform provider-specific schema to the common normalized schema.
+
+        ``loss_report`` accumulates counts of dropped/unhandled content so the
+        export loop can surface a single summary at the end. When None, providers
+        construct a throwaway local report (so calling normalize_conversation in
+        isolation, e.g. from tests, doesn't crash).
+        """
+
+    def check_drift(self) -> list[dict]:
+        """Probe the live API and assert the normalizer's load-bearing fields.
+
+        Returns a list of ``drift_finding`` dicts. The default is a no-op
+        (used by providers with no remote schema risk, e.g. local Claude Code
+        transcripts); web-API providers override this. See FUTURE.md §10.
+        """
+        return []
 
     # ------------------------------------------------------------------
     # Concrete helpers
     # ------------------------------------------------------------------
 
+    def _pace(self) -> None:
+        """Sleep so consecutive requests are at least REQUEST_DELAY apart (±25% jitter).
+
+        getattr fallback: tests construct providers via ``__new__``, skipping
+        ``__init__`` — those run unpaced.
+        """
+        delay = getattr(self, "_request_delay", 0)
+        if delay <= 0:
+            return
+        target = delay * random.uniform(0.75, 1.25)
+        last = getattr(self, "_last_request_at", None)
+        if last is not None:
+            wait = target - (time.monotonic() - last)
+            if wait > 0:
+                time.sleep(wait)
+        self._last_request_at = time.monotonic()
+
     def fetch_all_conversations(self, since: datetime | None = None) -> list[dict]:
         """Fetch every conversation, handling pagination automatically.
 
@@ -201,6 +315,7 @@ class BaseProvider(ABC):
 
         while attempt <= MAX_RETRIES:
             attempt += 1
+            self._pace()
             start = time.monotonic()
 
             try:

src/providers/claude.py

@@ -5,7 +5,25 @@ import os
 
 from curl_cffi import requests as curl_requests
 
-from src.providers.base import BaseProvider, ProviderError
+from src.blocks import (
+    UNKNOWN_REASON_EXTRACTION_FAILED,
+    UNKNOWN_REASON_UNKNOWN_TYPE,
+    make_image_placeholder,
+    make_text_block,
+    make_thinking_block,
+    make_tool_result_block,
+    make_tool_use_block,
+    make_unknown_block,
+)
+from src.loss_report import LossReport
+from src.providers.base import (
+    BaseProvider,
+    DRIFT_ERROR,
+    DRIFT_OK,
+    DRIFT_WARN,
+    ProviderError,
+    drift_finding,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -161,8 +179,9 @@ class ClaudeProvider(BaseProvider):
 
         return data
 
-    def normalize_conversation(self, raw: dict) -> dict:
+    def normalize_conversation(self, raw: dict, loss_report: LossReport | None = None) -> dict:
         """Transform Claude raw schema to the common normalized schema."""
+        report = loss_report if loss_report is not None else LossReport()
         conv_id = raw.get("uuid") or raw.get("id", "")
         title = raw.get("name") or raw.get("title") or "Untitled"
         created_at = raw.get("created_at") or raw.get("create_time") or ""
@@ -178,40 +197,37 @@ class ClaudeProvider(BaseProvider):
 
         # Messages
         raw_messages = raw.get("chat_messages") or raw.get("messages") or []
-        messages = []
+        messages: list[dict] = []
 
         for msg in raw_messages:
             role = _map_role(msg.get("sender") or msg.get("role", ""))
             if not role:
                 continue
 
-            # Content can be a string or a list of content blocks
+            content_raw = msg.get("content") if "content" in msg else msg.get("text", "")
-            content_raw = msg.get("content") or msg.get("text") or ""
+            blocks = _extract_claude_blocks(content_raw, conv_id, report)
-            content, skipped_types = _extract_claude_text(content_raw, conv_id)
-
-            for ctype in skipped_types:
-                logger.warning(
-                    "[claude] Skipping %s content in conversation %s "
-                    "— rich content not yet supported (see FUTURE.md)",
-                    ctype,
-                    conv_id[:8],
-                )
 
             timestamp = msg.get("created_at") or msg.get("timestamp") or None
 
-            if content is None:
+            if not blocks:
                 logger.debug("[claude] Skipping empty message in conversation %s", conv_id[:8])
                 continue
 
+            content_type = "text"
+
             messages.append(
                 {
                     "role": role,
-                    "content": content,
+                    "content_type": content_type,
-                    "content_type": "text",
                     "timestamp": timestamp,
+                    "blocks": blocks,
                 }
             )
 
+        for _ in messages:
+            report.record_message()
+        report.record_conversation()
+
         return {
             "id": conv_id,
             "title": title,
@@ -223,6 +239,70 @@ class ClaudeProvider(BaseProvider):
             "messages": messages,
         }
 
+    def check_drift(self) -> list[dict]:
+        """Probe one listing page + one conversation; assert load-bearing fields.
+
+        See FUTURE.md §10. Real Claude messages are flat ``text``/``sender``;
+        the rich ``content`` block path is dormant, so its activation (content
+        becoming a list) is itself a drift event worth flagging. ``attachments``
+        / ``files`` are ignored by the normalizer — non-empty values are a
+        silent-loss risk.
+        """
+        findings: list[dict] = []
+        page = self.list_conversations(offset=0, limit=5)
+        if not page:
+            return [drift_finding(DRIFT_WARN, "listing",
+                                  "empty listing — cannot verify shape")]
+
+        item = page[0]
+        if not (item.get("uuid") or item.get("id")):
+            findings.append(drift_finding(DRIFT_ERROR, "listing", "summary missing uuid/id"))
+        if not (item.get("name") or item.get("title")):
+            findings.append(drift_finding(DRIFT_WARN, "listing", "no name/title on summary"))
+        if not (item.get("updated_at") or item.get("update_time")):
+            findings.append(drift_finding(DRIFT_WARN, "listing", "no updated_at on summary"))
+
+        conv_id = item.get("uuid") or item.get("id")
+        if not conv_id:
+            return findings or [drift_finding(DRIFT_ERROR, "listing", "no id to fetch")]
+        raw = self.get_conversation(conv_id)
+        if not (raw.get("uuid") or raw.get("id")):
+            findings.append(drift_finding(DRIFT_ERROR, "detail", "no uuid/id on detail"))
+        msgs = raw.get("chat_messages") or raw.get("messages")
+        if not isinstance(msgs, list) or not msgs:
+            findings.append(drift_finding(
+                DRIFT_ERROR, "detail", "chat_messages missing/empty — 0 messages"))
+            return findings
+
+        saw_sender = saw_text = False
+        flagged_list = flagged_attach = False
+        for m in msgs:
+            if m.get("sender") or m.get("role"):
+                saw_sender = True
+            content = m.get("content")
+            if isinstance(content, list) and not flagged_list:
+                findings.append(drift_finding(
+                    DRIFT_WARN, "content",
+                    "message.content is now a LIST — Claude switched to typed blocks; "
+                    "the dormant _dispatch_claude_block path is now live, verify rendering"))
+                flagged_list = True
+            if m.get("text") or content:
+                saw_text = True
+            if (m.get("attachments") or m.get("files")) and not flagged_attach:
+                findings.append(drift_finding(
+                    DRIFT_WARN, "attachments",
+                    "message has non-empty attachments/files — the normalizer ignores "
+                    "these (silent loss); add handling"))
+                flagged_attach = True
+
+        if not saw_sender:
+            findings.append(drift_finding(DRIFT_ERROR, "messages", "no sender/role on any message"))
+        if not saw_text:
+            findings.append(drift_finding(DRIFT_WARN, "messages", "no text/content found"))
+        if not findings:
+            findings.append(drift_finding(DRIFT_OK, "schema", "all load-bearing fields present"))
+        return findings
+
 
 # ---------------------------------------------------------------------------
 # Internal helpers
@@ -242,43 +322,134 @@ def _map_role(sender: str) -> str | None:
     return mapping.get(sender.lower()) if sender else None
 
 
-def _extract_claude_text(
+def _extract_claude_blocks(
-    content: str | list | dict, conv_id: str
+    content: str | list | dict | None, conv_id: str, report: LossReport
-) -> tuple[str | None, list[str]]:
+) -> list[dict]:
-    """Extract plain text from a Claude content field.
+    """Extract typed blocks from a Claude content field.
 
-    Returns:
+    Defensive dispatch — zero observed cases of rich Claude content in the
-        (text_or_None, list_of_skipped_content_types)
+    user's archive at planning time, so this is theory-only. Real shapes
+    will be locked in v0.4.1 once captured. Any unrecognised block type
+    surfaces as an `unknown` block + WARNING + tally.
     """
-    skipped: list[str] = []
+    if content is None:
+        return []
 
     if isinstance(content, str):
-        text = content.strip()
+        block = make_text_block(content)
-        return (text if text else None), skipped
+        return [block] if block else []
 
     if isinstance(content, list):
-        parts: list[str] = []
+        blocks: list[dict] = []
-        for block in content:
+        for item in content:
-            if isinstance(block, str):
+            if isinstance(item, str):
-                parts.append(block)
+                block = make_text_block(item)
-            elif isinstance(block, dict):
+                if block:
-                btype = block.get("type", "text")
+                    blocks.append(block)
-                if btype == "text":
+            elif isinstance(item, dict):
-                    t = block.get("text", "").strip()
+                blocks.extend(_dispatch_claude_block(item, conv_id, report))
-                    if t:
+        return blocks
-                        parts.append(t)
-                else:
-                    skipped.append(btype)
-        text = "\n".join(parts).strip()
-        return (text if text else None), skipped
 
     if isinstance(content, dict):
-        btype = content.get("type", "text")
+        return _dispatch_claude_block(content, conv_id, report)
-        if btype == "text":
-            text = content.get("text", "").strip()
-            return (text if text else None), skipped
-        else:
-            skipped.append(btype)
-            return None, skipped
 
-    return None, skipped
+    return []
+
+
+def _dispatch_claude_block(block: dict, conv_id: str, report: LossReport) -> list[dict]:
+    """Translate one raw Claude content block into normalized blocks."""
+    btype = block.get("type", "text")
+
+    if btype == "text":
+        block_obj = make_text_block(block.get("text", "") or "")
+        return [block_obj] if block_obj else []
+
+    if btype == "thinking":
+        # Claude extended-thinking blocks may use 'thinking' or 'text' field.
+        text = block.get("thinking") or block.get("text") or ""
+        block_obj = make_thinking_block(text)
+        return [block_obj] if block_obj else []
+
+    if btype == "tool_use":
+        return [
+            make_tool_use_block(
+                name=block.get("name", "") or "",
+                input_data=block.get("input"),
+                tool_id=block.get("id"),
+            )
+        ]
+
+    if btype == "tool_result":
+        # ``content`` may be a string or a list of nested blocks (recursive).
+        nested = block.get("content")
+        output = _flatten_tool_result_content(nested, conv_id, report)
+        return [
+            make_tool_result_block(
+                output=output,
+                tool_name=None,
+                is_error=bool(block.get("is_error")),
+            )
+        ]
+
+    if btype == "image":
+        # Source shape is unverified; try the most likely fields.
+        source = block.get("source") or {}
+        ref = ""
+        if isinstance(source, dict):
+            ref = (
+                source.get("file_uuid")
+                or source.get("media_type")
+                or source.get("url")
+                or ""
+            )
+        return [make_image_placeholder(ref=ref or "(unknown)", source="user_upload")]
+
+    # Unknown block type
+    keys = list(block.keys())
+    logger.warning(
+        "[claude] Unknown block type %r in conversation %s "
+        "— see plan §Data-loss visibility (rendering as unknown block)",
+        btype,
+        conv_id[:8],
+    )
+    report.record_unknown(btype or "?")
+    return [
+        make_unknown_block(
+            raw_type=btype or "?",
+            observed_keys=keys,
+            reason=UNKNOWN_REASON_UNKNOWN_TYPE,
+        )
+    ]
+
+
+def _flatten_tool_result_content(
+    nested: object, conv_id: str, report: LossReport
+) -> str:
+    """Flatten Claude tool_result content (string OR list of nested blocks) to text.
+
+    Recurses into nested text blocks; any non-text nested block becomes a
+    visible inline marker so non-text content isn't silently dropped.
+    """
+    if nested is None:
+        return ""
+    if isinstance(nested, str):
+        return nested
+    if isinstance(nested, list):
+        chunks: list[str] = []
+        for item in nested:
+            if isinstance(item, str):
+                chunks.append(item)
+            elif isinstance(item, dict):
+                btype = item.get("type", "text")
+                if btype == "text":
+                    chunks.append(item.get("text", "") or "")
+                else:
+                    keys = list(item.keys())[:10]
+                    report.record_extraction_failure(f"tool_result.{btype}")
+                    chunks.append(
+                        f"[Unsupported nested {btype} block; keys={keys}]"
+                    )
+        return "\n".join(c for c in chunks if c)
+    if isinstance(nested, dict):
+        return _flatten_tool_result_content([nested], conv_id, report)
+    return str(nested)

src/providers/claude_code.py

@@ -0,0 +1,476 @@
+"""Claude Code session provider — archives local agent transcripts.
+
+Reads JSONL session files from ``~/.claude/projects/<munged-cwd>/<uuid>.jsonl``
+(override with ``CLAUDE_CODE_DIR``). No tokens, no rate limits, no ToS risk —
+but the data lives in single files Claude Code may clean up, and it contains
+deliverables (reviews, plans, analyses) that exist nowhere else.
+
+Rendering follows the EXPORTER_HIDDEN_CONTENT policy (decided 2026-06-12):
+prose-only by default. User prompts and assistant text are kept; tool_use /
+tool_result traffic is collapsed to one grouped placeholder per activity run
+(measured: dialogue prose is ~4% of session bytes); thinking blocks are
+dropped (counted in the run summary, no placeholder). ``full`` keeps
+everything.
+
+Record types in a session file: ``user`` / ``assistant`` carry the dialogue
+(Anthropic-style ``message.content`` block arrays); ``ai-title`` carries the
+evolving session title (last one wins); ``last-prompt``,
+``file-history-snapshot``, ``attachment``, ``permission-mode``, ``system``
+are harness records and are skipped. Records flagged ``isSidechain`` are
+subagent transcripts; ``isMeta`` are harness-generated user records — both
+skipped.
+"""
+
+import json
+import logging
+import os
+import re
+from collections import Counter
+from datetime import datetime, timezone
+from pathlib import Path
+
+from src.blocks import (
+    COLLAPSED_KIND_TOOL_DUMP,
+    UNKNOWN_REASON_UNKNOWN_TYPE,
+    make_collapsed_block,
+    make_image_placeholder,
+    make_text_block,
+    make_thinking_block,
+    make_tool_result_block,
+    make_tool_use_block,
+    make_unknown_block,
+)
+from src.loss_report import LossReport
+from src.providers.base import (
+    BaseProvider,
+    HIDDEN_CONTENT_FULL,
+    ProviderError,
+    VALID_HIDDEN_CONTENT_POLICIES,
+    resolve_hidden_content_policy,
+)
+
+logger = logging.getLogger(__name__)
+
+DEFAULT_PROJECTS_DIR = "~/.claude/projects"
+
+# Harness-injected tags inside user message text. Stripped so exports contain
+# the dialogue, not the CLI plumbing. A record that is nothing but tags
+# (e.g. a /model invocation) ends up empty and is skipped.
+_HARNESS_TAG_RE = re.compile(
+    r"<local-command-caveat>.*?</local-command-caveat>"
+    r"|<command-name>.*?</command-name>"
+    r"|<command-message>.*?</command-message>"
+    r"|<command-args>.*?</command-args>"
+    r"|<command-contents>.*?</command-contents>"
+    r"|<local-command-stdout>.*?</local-command-stdout>"
+    r"|<system-reminder>.*?</system-reminder>",
+    re.DOTALL,
+)
+
+# How many distinct tool names to list in a collapsed-activity placeholder.
+_TOOL_NAMES_SHOWN = 4
+
+
+def _strip_harness_noise(text: str) -> str:
+    if not isinstance(text, str):
+        return ""
+    return _HARNESS_TAG_RE.sub("", text).strip()
+
+
+class ClaudeCodeProvider(BaseProvider):
+    """Local-file provider over Claude Code session transcripts."""
+
+    provider_name = "claude-code"
+
+    def __init__(
+        self,
+        projects_dir: str | Path | None = None,
+        hidden_content: str | None = None,
+    ) -> None:
+        super().__init__()
+        self._projects_dir = Path(
+            projects_dir or os.getenv("CLAUDE_CODE_DIR", DEFAULT_PROJECTS_DIR)
+        ).expanduser()
+        self._hidden_content = (
+            hidden_content
+            if hidden_content in VALID_HIDDEN_CONTENT_POLICIES
+            else resolve_hidden_content_policy()
+        )
+        # conv_id → session file path, populated by _scan()
+        self._path_map: dict[str, Path] = {}
+
+    # ------------------------------------------------------------------
+    # BaseProvider interface
+    # ------------------------------------------------------------------
+
+    def list_conversations(self, offset: int = 0, limit: int = 100) -> list[dict]:
+        full = self._scan()
+        return full[offset : offset + limit]
+
+    def fetch_all_conversations(self, since: datetime | None = None) -> list[dict]:
+        convs = self._scan()
+        if since is not None:
+            since_aware = since if since.tzinfo else since.replace(tzinfo=timezone.utc)
+            convs = [
+                c for c in convs
+                if datetime.fromisoformat(c["updated_at"]) >= since_aware
+            ]
+        logger.info(
+            "[claude-code] Found %d session(s) under %s", len(convs), self._projects_dir
+        )
+        return convs
+
+    def get_conversation(self, conv_id: str) -> dict:
+        path = self._path_map.get(conv_id)
+        if path is None:
+            # Direct call without a prior listing (e.g. tests) — scan first.
+            self._scan()
+            path = self._path_map.get(conv_id)
+        if path is None or not path.exists():
+            raise ProviderError(
+                self.provider_name,
+                f"get_conversation({conv_id[:8]})",
+                FileNotFoundError(f"No session file for id {conv_id}"),
+            )
+
+        records: list[dict] = []
+        bad_lines = 0
+        for line in path.read_text(encoding="utf-8").splitlines():
+            line = line.strip()
+            if not line:
+                continue
+            try:
+                records.append(json.loads(line))
+            except json.JSONDecodeError:
+                bad_lines += 1
+        if bad_lines:
+            logger.warning(
+                "[claude-code] %s: skipped %d unparseable line(s)", path.name, bad_lines
+            )
+
+        mtime = datetime.fromtimestamp(path.stat().st_mtime, tz=timezone.utc)
+        return {
+            "id": conv_id,
+            "_path": str(path),
+            "_records": records,
+            # Listing and normalized updated_at must match, or the cache
+            # staleness comparison would re-export every session every run.
+            "_mtime_iso": mtime.isoformat(),
+        }
+
+    def normalize_conversation(self, raw: dict, loss_report: LossReport | None = None) -> dict:
+        report = loss_report if loss_report is not None else LossReport()
+        policy = getattr(self, "_hidden_content", None) or resolve_hidden_content_policy()
+        conv_id = raw.get("id") or ""
+        records: list[dict] = raw.get("_records") or []
+
+        title = _extract_title(records)
+        project = _extract_project(records, raw.get("_path"))
+        created_at = next(
+            (r.get("timestamp") for r in records if r.get("timestamp")), ""
+        )
+        updated_at = raw.get("_mtime_iso") or next(
+            (r.get("timestamp") for r in reversed(records) if r.get("timestamp")), ""
+        )
+
+        messages = _extract_messages(records, conv_id, report, policy)
+        for _ in messages:
+            report.record_message()
+        report.record_conversation()
+
+        return {
+            "id": conv_id,
+            "title": title,
+            "provider": self.provider_name,
+            "project": project,
+            "created_at": created_at or "",
+            "updated_at": updated_at or "",
+            "message_count": len(messages),
+            "messages": messages,
+        }
+
+    # ------------------------------------------------------------------
+    # Scanning
+    # ------------------------------------------------------------------
+
+    def _scan(self) -> list[dict]:
+        if not self._projects_dir.is_dir():
+            logger.warning(
+                "[claude-code] Projects directory %s does not exist", self._projects_dir
+            )
+            return []
+
+        convs: list[dict] = []
+        for proj_dir in sorted(p for p in self._projects_dir.iterdir() if p.is_dir()):
+            for session_file in sorted(proj_dir.glob("*.jsonl")):
+                try:
+                    stat = session_file.stat()
+                except OSError:
+                    continue
+                if stat.st_size == 0:
+                    continue
+                conv_id = session_file.stem
+                self._path_map[conv_id] = session_file
+                title, project, created = _read_session_meta(session_file)
+                convs.append(
+                    {
+                        "id": conv_id,
+                        "title": title,
+                        "project": project,
+                        # The --project filter and dry-run table read the
+                        # listing dict, not the normalized conversation.
+                        "_project_name": project,
+                        "created_at": created,
+                        "updated_at": datetime.fromtimestamp(
+                            stat.st_mtime, tz=timezone.utc
+                        ).isoformat(),
+                        "_path": str(session_file),
+                        "_project_dir": proj_dir.name,
+                    }
+                )
+        return convs
+
+
+# ---------------------------------------------------------------------------
+# Internal helpers
+# ---------------------------------------------------------------------------
+
+
+def _read_session_meta(path: Path) -> tuple[str, str | None, str]:
+    """Light single-pass scan for listing metadata: (title, project, created_at).
+
+    Substring guards keep this cheap — only candidate lines are JSON-parsed.
+    The full-fidelity extraction happens later in normalize_conversation.
+    """
+    title = ""
+    project: str | None = None
+    created = ""
+    try:
+        with path.open(encoding="utf-8") as fh:
+            for line in fh:
+                if '"ai-title"' in line:
+                    try:
+                        rec = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    if rec.get("type") == "ai-title" and rec.get("aiTitle"):
+                        title = str(rec["aiTitle"])  # last one wins
+                    continue
+                if (not created or project is None) and (
+                    '"timestamp"' in line or '"cwd"' in line
+                ):
+                    try:
+                        rec = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    if not created and rec.get("timestamp"):
+                        created = str(rec["timestamp"])
+                    if project is None and rec.get("cwd"):
+                        project = Path(rec["cwd"]).name or None
+    except OSError as e:
+        logger.warning("[claude-code] Could not read %s: %s", path, e)
+    return title, project, created
+
+
+def _extract_title(records: list[dict]) -> str:
+    """Last ai-title record wins; fall back to the first real user prompt."""
+    title = ""
+    for rec in records:
+        if rec.get("type") == "ai-title" and rec.get("aiTitle"):
+            title = str(rec["aiTitle"])
+    if title:
+        return title
+
+    for rec in records:
+        if rec.get("type") != "user" or rec.get("isMeta") or rec.get("isSidechain"):
+            continue
+        content = (rec.get("message") or {}).get("content")
+        if isinstance(content, str):
+            text = _strip_harness_noise(content)
+        elif isinstance(content, list):
+            text = " ".join(
+                _strip_harness_noise(item.get("text", ""))
+                for item in content
+                if isinstance(item, dict) and item.get("type") == "text"
+            ).strip()
+        else:
+            text = ""
+        if text:
+            return text[:80]
+    return "Untitled session"
+
+
+def _extract_project(records: list[dict], path: str | None) -> str | None:
+    """Project = basename of the session's working directory."""
+    for rec in records:
+        cwd = rec.get("cwd")
+        if cwd:
+            name = Path(cwd).name
+            if name:
+                return name
+    # Fallback: the munged directory name (cannot be reliably de-munged
+    # because '-' is both the path separator and a legal name character).
+    if path:
+        return Path(path).parent.name.lstrip("-") or None
+    return None
+
+
+def _extract_messages(
+    records: list[dict], conv_id: str, report: LossReport, policy: str
+) -> list[dict]:
+    messages: list[dict] = []
+    # Pending collapsed tool activity: name → call count, plus total bytes.
+    pending_tools: Counter = Counter()
+    pending_bytes = 0
+
+    def flush_pending() -> None:
+        nonlocal pending_bytes
+        if not pending_tools:
+            return
+        shown = ", ".join(
+            f"{name} ×{count}" for name, count in pending_tools.most_common(_TOOL_NAMES_SHOWN)
+        )
+        if len(pending_tools) > _TOOL_NAMES_SHOWN:
+            shown += ", …"
+        calls = sum(pending_tools.values())
+        block = make_collapsed_block(
+            origin=f"{calls} calls: {shown}",
+            content_type="tool_activity",
+            size_bytes=pending_bytes,
+            kind=COLLAPSED_KIND_TOOL_DUMP,
+        )
+        if messages and messages[-1]["role"] == "assistant":
+            messages[-1]["blocks"].append(block)
+        else:
+            messages.append(
+                {"role": "tool", "content_type": "text", "timestamp": None, "blocks": [block]}
+            )
+        pending_tools.clear()
+        pending_bytes = 0
+
+    for rec in records:
+        if rec.get("type") not in ("user", "assistant"):
+            continue
+        if rec.get("isSidechain") or rec.get("isMeta"):
+            continue
+
+        msg = rec.get("message") or {}
+        role = msg.get("role") or rec["type"]
+        content = msg.get("content")
+        blocks: list[dict] = []
+        # This record's own tool traffic — merged into pending only after the
+        # record's message is appended, so the placeholder lands on it (not on
+        # the previous message).
+        local_tools: Counter = Counter()
+        local_bytes = 0
+
+        if isinstance(content, str):
+            text = _strip_harness_noise(content)
+            block = make_text_block(text)
+            if block:
+                blocks.append(block)
+        elif isinstance(content, list):
+            for item in content:
+                if not isinstance(item, dict):
+                    continue
+                item_type = item.get("type", "")
+                if item_type == "text":
+                    block = make_text_block(_strip_harness_noise(item.get("text", "")))
+                    if block:
+                        blocks.append(block)
+                elif item_type in ("thinking", "redacted_thinking"):
+                    if policy == HIDDEN_CONTENT_FULL:
+                        block = make_thinking_block(
+                            item.get("thinking") or item.get("text") or ""
+                        )
+                        if block:
+                            blocks.append(block)
+                    else:
+                        # Decision 2026-06-12: thinking is dropped without a
+                        # placeholder, but stays visible in the run summary.
+                        report.record_collapsed(
+                            "thinking", len(json.dumps(item, default=str))
+                        )
+                elif item_type == "tool_use":
+                    if policy == HIDDEN_CONTENT_FULL:
+                        blocks.append(
+                            make_tool_use_block(
+                                item.get("name", ""), item.get("input"), item.get("id")
+                            )
+                        )
+                    else:
+                        name = item.get("name") or "tool"
+                        size = len(json.dumps(item, default=str))
+                        local_tools[name] += 1
+                        local_bytes += size
+                        report.record_collapsed(name, size)
+                elif item_type == "tool_result":
+                    if policy == HIDDEN_CONTENT_FULL:
+                        blocks.append(
+                            make_tool_result_block(
+                                _stringify_tool_result(item.get("content")),
+                                is_error=bool(item.get("is_error")),
+                            )
+                        )
+                    else:
+                        size = len(json.dumps(item, default=str))
+                        local_bytes += size
+                        report.record_collapsed("tool_result", size)
+                elif item_type == "image":
+                    blocks.append(
+                        make_image_placeholder(ref="embedded image", source="user_upload")
+                    )
+                else:
+                    logger.warning(
+                        "[claude-code] Unknown content block type %r in session %s",
+                        item_type,
+                        conv_id[:8],
+                    )
+                    report.record_unknown(f"claude-code.{item_type or '?'}")
+                    blocks.append(
+                        make_unknown_block(
+                            raw_type=f"claude-code.{item_type or '?'}",
+                            observed_keys=list(item.keys()),
+                            reason=UNKNOWN_REASON_UNKNOWN_TYPE,
+                        )
+                    )
+
+        if not blocks:
+            # Tool-result-only records (and similar): traffic accumulates and
+            # is attached to the message that initiated it.
+            pending_tools.update(local_tools)
+            pending_bytes += local_bytes
+            continue
+
+        # Attach previous records' tool activity to the previous message
+        # before starting a new one, so the placeholder lands between the
+        # dialogue turns it actually occurred between.
+        flush_pending()
+        messages.append(
+            {
+                "role": role,
+                "content_type": "text",
+                "timestamp": rec.get("timestamp"),
+                "blocks": blocks,
+            }
+        )
+        pending_tools.update(local_tools)
+        pending_bytes += local_bytes
+
+    flush_pending()
+    return messages
+
+
+def _stringify_tool_result(content) -> str:
+    """tool_result content may be a string or a list of text blocks."""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = []
+        for item in content:
+            if isinstance(item, dict) and item.get("type") == "text":
+                parts.append(item.get("text", ""))
+            else:
+                parts.append(json.dumps(item, default=str))
+        return "\n".join(parts)
+    return json.dumps(content, default=str) if content is not None else ""

src/utils.py

@@ -50,7 +50,7 @@ def build_export_path(
         created_at: ISO8601 creation timestamp (used for year folder).
         filename: Already-generated filename from generate_filename().
         structure: OUTPUT_STRUCTURE value. One of:
-            "provider/project/year" (default)
+            "provider/project/year" (default) — project and year combined, e.g. no-project.2025/
             "provider/project"
             "provider/year"
 
@@ -64,14 +64,14 @@ def build_export_path(
     parts: list[str] = [provider]
 
     if structure == "provider/project/year":
-        parts += [project_slug, year]
+        parts += [f"{project_slug}.{year}"]
     elif structure == "provider/project":
         parts += [project_slug]
     elif structure == "provider/year":
         parts += [year]
     else:
         # Unknown structure — fall back to default
-        parts += [project_slug, year]
+        parts += [f"{project_slug}.{year}"]
 
     return base_dir.joinpath(*parts) / filename
 

tests/fixtures/chatgpt_conversation.json

@@ -8,12 +8,30 @@
     "node-root": {
       "id": "node-root",
       "parent": null,
-      "children": ["node-1"],
+      "children": ["node-uec"],
       "message": null
     },
+    "node-uec": {
+      "id": "node-uec",
+      "parent": "node-root",
+      "children": ["node-1"],
+      "message": {
+        "id": "node-uec",
+        "author": {"role": "user"},
+        "create_time": null,
+        "content": {
+          "content_type": "user_editable_context",
+          "user_profile": "Preferred name: Jesse",
+          "user_instructions": "The user provided the additional info about how they would like you to respond:\n```Always cite sources.```"
+        },
+        "metadata": {
+          "is_visually_hidden_from_conversation": true
+        }
+      }
+    },
     "node-1": {
       "id": "node-1",
-      "parent": "node-root",
+      "parent": "node-uec",
       "children": ["node-2"],
       "message": {
         "id": "node-1",
@@ -28,7 +46,7 @@
     "node-2": {
       "id": "node-2",
       "parent": "node-1",
-      "children": ["node-3"],
+      "children": ["node-mm-user"],
       "message": {
         "id": "node-2",
         "author": {"role": "assistant"},
@@ -39,19 +57,140 @@
         }
       }
     },
-    "node-3": {
+    "node-mm-user": {
-      "id": "node-3",
+      "id": "node-mm-user",
       "parent": "node-2",
-      "children": [],
+      "children": ["node-mm-assistant"],
       "message": {
-        "id": "node-3",
+        "id": "node-mm-user",
         "author": {"role": "user"},
         "create_time": 1704067300.0,
         "content": {
-          "content_type": "image_asset_pointer",
+          "content_type": "multimodal_text",
-          "parts": [{"content_type": "image_asset_pointer", "asset_pointer": "file://some-image"}]
+          "parts": [
+            {"content_type": "audio_transcription", "text": "What is the capital of France?", "direction": "in", "decoding_id": null},
+            {"content_type": "real_time_user_audio_video_asset_pointer", "frames_asset_pointers": [], "video_container_asset_pointer": null, "audio_asset_pointer": {"content_type": "audio_asset_pointer", "asset_pointer": "sediment://file_user001", "size_bytes": 50000, "format": "wav", "metadata": {"start": 0.0, "end": 2.5}}, "audio_start_timestamp": 1.0}
+          ]
+        },
+        "metadata": {"voice_mode_message": true}
+      }
+    },
+    "node-mm-assistant": {
+      "id": "node-mm-assistant",
+      "parent": "node-mm-user",
+      "children": ["node-mm-user-rev"],
+      "message": {
+        "id": "node-mm-assistant",
+        "author": {"role": "assistant"},
+        "create_time": 1704067305.0,
+        "content": {
+          "content_type": "multimodal_text",
+          "parts": [
+            {"content_type": "audio_transcription", "text": "The capital of France is Paris.", "direction": "out", "decoding_id": null},
+            {"content_type": "audio_asset_pointer", "asset_pointer": "sediment://file_assistant001", "size_bytes": 80000, "format": "wav", "metadata": {"start": 0.0, "end": 3.2}}
+          ]
         }
       }
+    },
+    "node-mm-user-rev": {
+      "id": "node-mm-user-rev",
+      "parent": "node-mm-assistant",
+      "children": ["node-image-only"],
+      "message": {
+        "id": "node-mm-user-rev",
+        "author": {"role": "user"},
+        "create_time": 1704067400.0,
+        "content": {
+          "content_type": "multimodal_text",
+          "parts": [
+            {"content_type": "real_time_user_audio_video_asset_pointer", "frames_asset_pointers": [], "video_container_asset_pointer": null, "audio_asset_pointer": {"content_type": "audio_asset_pointer", "asset_pointer": "sediment://file_user002", "size_bytes": 30000, "format": "wav", "metadata": {"start": 0.0, "end": 1.5}}, "audio_start_timestamp": 5.0},
+            {"content_type": "audio_transcription", "text": "Tell me more please.", "direction": "in", "decoding_id": null}
+          ]
+        }
+      }
+    },
+    "node-image-only": {
+      "id": "node-image-only",
+      "parent": "node-mm-user-rev",
+      "children": ["node-exec-output"],
+      "message": {
+        "id": "node-image-only",
+        "author": {"role": "user"},
+        "create_time": 1704067500.0,
+        "content": {
+          "content_type": "multimodal_text",
+          "parts": [
+            {"content_type": "image_asset_pointer", "asset_pointer": "file-service://image001"}
+          ]
+        }
+      }
+    },
+    "node-exec-output": {
+      "id": "node-exec-output",
+      "parent": "node-image-only",
+      "children": ["node-exec-output-empty"],
+      "message": {
+        "id": "node-exec-output",
+        "author": {"role": "tool", "name": "container.exec", "metadata": {}},
+        "create_time": 1704067600.0,
+        "content": {
+          "content_type": "execution_output",
+          "text": "Hello from container.exec\nLine 2 of output"
+        },
+        "metadata": {
+          "aggregate_result": {"status": "success", "messages": []},
+          "reasoning_title": "Reading skill documentation"
+        }
+      }
+    },
+    "node-exec-output-empty": {
+      "id": "node-exec-output-empty",
+      "parent": "node-exec-output",
+      "children": ["node-system-error"],
+      "message": {
+        "id": "node-exec-output-empty",
+        "author": {"role": "tool", "name": "python", "metadata": {}},
+        "create_time": 1704067610.0,
+        "content": {
+          "content_type": "execution_output",
+          "text": ""
+        },
+        "metadata": {}
+      }
+    },
+    "node-system-error": {
+      "id": "node-system-error",
+      "parent": "node-exec-output-empty",
+      "children": ["node-tether-spinner"],
+      "message": {
+        "id": "node-system-error",
+        "author": {"role": "tool", "name": "web", "metadata": {}},
+        "create_time": 1704067620.0,
+        "content": {
+          "content_type": "system_error",
+          "name": "tool_error",
+          "text": "Error: Error from browse service: Error calling browse service: 503"
+        },
+        "metadata": {}
+      }
+    },
+    "node-tether-spinner": {
+      "id": "node-tether-spinner",
+      "parent": "node-system-error",
+      "children": [],
+      "message": {
+        "id": "node-tether-spinner",
+        "author": {"role": "tool", "name": "file_search", "metadata": {}},
+        "create_time": 1704067630.0,
+        "content": {
+          "content_type": "tether_browsing_display",
+          "result": "",
+          "summary": "",
+          "assets": null,
+          "tether_id": null
+        },
+        "metadata": {"command": "spinner", "status": "running"}
+      }
     }
   }
 }

tests/fixtures/claude_conversation.json

@@ -30,6 +30,15 @@
       "sender": "human",
       "created_at": "2024-06-10T14:45:00.000Z",
       "content": "Thank you, that helped!"
+    },
+    {
+      "uuid": "msg-004",
+      "sender": "human",
+      "created_at": "2024-06-10T14:50:00.000Z",
+      "content": [
+        {"type": "text", "text": "What about this image?"},
+        {"type": "image", "source": {"file_uuid": "claude-image-uuid-1", "media_type": "image/png"}}
+      ]
     }
   ]
 }

tests/test_cache.py

@@ -151,3 +151,106 @@ class TestGetNewOrUpdated:
         convs = [{"id": "a", "updated_at": "2024-06-01T00:00:00Z"}]
         result = tmp_cache.get_new_or_updated("claude", convs)
         assert len(result) == 1
+
+    def test_force_returns_all_cached_and_unchanged(self, tmp_cache):
+        tmp_cache.mark_exported("claude", "a", {"updated_at": "2024-01-01T00:00:00Z"})
+        convs = [
+            {"id": "a", "updated_at": "2024-01-01T00:00:00Z"},  # cached, unchanged
+            {"id": "b", "updated_at": "2024-01-02T00:00:00Z"},  # new
+        ]
+        assert len(tmp_cache.get_new_or_updated("claude", convs)) == 1
+        assert len(tmp_cache.get_new_or_updated("claude", convs, force=True)) == 2
+
+    def test_force_orders_least_recently_exported_first(self, tmp_cache):
+        # 'a' was exported long ago; 'b' just now; 'c' never.
+        tmp_cache.mark_exported("claude", "a", {"updated_at": "2024-01-01T00:00:00Z"})
+        tmp_cache._data["claude"]["a"]["exported_at"] = "2024-01-01T00:00:00Z"
+        tmp_cache.mark_exported("claude", "b", {"updated_at": "2024-01-01T00:00:00Z"})
+        tmp_cache._data["claude"]["b"]["exported_at"] = "2026-06-12T00:00:00Z"
+        convs = [{"id": "a"}, {"id": "b"}, {"id": "c"}]
+        ordered = [c["id"] for c in tmp_cache.get_new_or_updated("claude", convs, force=True)]
+        # never-exported ('c', exported_at "") and oldest ('a') come before 'b'
+        assert ordered == ["c", "a", "b"]
+
+    def test_force_capped_run_makes_progress(self, tmp_cache):
+        """The bug: force + cap repeated the same head every run. Now each
+        capped run re-exports the next-oldest batch and converges."""
+        convs = [{"id": f"c{i}", "updated_at": "2024-01-01T00:00:00Z"} for i in range(5)]
+        seen = set()
+        for _ in range(3):  # cap=2 over 3 runs should cover all 5
+            batch = tmp_cache.get_new_or_updated("claude", convs, force=True)[:2]
+            for c in batch:
+                seen.add(c["id"])
+                tmp_cache.mark_exported("claude", c["id"], {"file_path": f"/{c['id']}.md"})
+        assert seen == {f"c{i}" for i in range(5)}
+
+    def test_campaign_excludes_already_refreshed(self, tmp_cache):
+        """With a campaign stamp, conversations re-exported during the campaign
+        (exported_at >= stamp) drop out, so the remaining count shrinks to 0."""
+        convs = [{"id": f"c{i}"} for i in range(5)]
+        campaign = "2026-06-12T00:00:00+00:00"
+        # Nothing exported yet → all 5 are candidates.
+        assert len(tmp_cache.get_new_or_updated("claude", convs, force=True,
+                                                campaign_at=campaign)) == 5
+        # Re-export 2 "during" the campaign (after the stamp).
+        for cid in ("c0", "c1"):
+            tmp_cache.mark_exported("claude", cid, {"file_path": f"/{cid}.md"})
+            tmp_cache._data["claude"][cid]["exported_at"] = "2026-06-12T01:00:00+00:00"
+        remaining = tmp_cache.get_new_or_updated("claude", convs, force=True,
+                                                 campaign_at=campaign)
+        assert {c["id"] for c in remaining} == {"c2", "c3", "c4"}
+        # Finish them → none left.
+        for cid in ("c2", "c3", "c4"):
+            tmp_cache.mark_exported("claude", cid, {"file_path": f"/{cid}.md"})
+            tmp_cache._data["claude"][cid]["exported_at"] = "2026-06-12T02:00:00+00:00"
+        assert tmp_cache.get_new_or_updated("claude", convs, force=True,
+                                            campaign_at=campaign) == []
+
+    def test_force_campaign_marker_roundtrip(self, tmp_cache):
+        assert tmp_cache.get_force_campaign() is None
+        tmp_cache.set_force_campaign("2026-06-12T00:00:00+00:00")
+        assert tmp_cache.get_force_campaign() == "2026-06-12T00:00:00+00:00"
+        # Survives reload, and is not mistaken for a provider by stats().
+        reopened = Cache(tmp_cache._dir)
+        assert reopened.get_force_campaign() == "2026-06-12T00:00:00+00:00"
+        assert "force_campaign_at" not in reopened.stats()
+        tmp_cache.clear_force_campaign()
+        assert tmp_cache.get_force_campaign() is None
+
+
+class TestReexportPreservesJoplinLink:
+    """A re-export must not drop the Joplin note link, or re-syncing would
+    create duplicate notes instead of updating the existing ones."""
+
+    def test_joplin_fields_carried_across_reexport(self, tmp_cache):
+        tmp_cache.mark_exported("chatgpt", "c1", {
+            "title": "T", "updated_at": "2024-01-01T00:00:00Z", "file_path": "/x.md",
+        })
+        tmp_cache.mark_joplin_synced("chatgpt", "c1", "note-123")
+        tmp_cache.set_joplin_resources("chatgpt", "c1", {"media/a.png": "res-1"})
+
+        # Re-export the same conversation (new render, new file_path).
+        tmp_cache.mark_exported("chatgpt", "c1", {
+            "title": "T", "updated_at": "2024-01-01T00:00:00Z", "file_path": "/x2.md",
+        })
+
+        entry = tmp_cache.get_all_entries("chatgpt")["c1"]
+        assert entry["joplin_note_id"] == "note-123"
+        assert entry["joplin_resources"] == {"media/a.png": "res-1"}
+        assert entry["file_path"] == "/x2.md"
+
+    def test_reexport_marks_note_for_update_not_create(self, tmp_cache):
+        tmp_cache.mark_exported("chatgpt", "c1", {
+            "updated_at": "2024-01-01T00:00:00Z", "file_path": "/x.md",
+        })
+        tmp_cache.mark_joplin_synced("chatgpt", "c1", "note-123")
+        # Nothing pending right after sync.
+        assert tmp_cache.get_joplin_pending("chatgpt") == []
+        # Re-export bumps exported_at past joplin_synced_at → pending for update,
+        # carrying the existing note id so the sync updates rather than creates.
+        tmp_cache.mark_exported("chatgpt", "c1", {
+            "updated_at": "2024-01-01T00:00:00Z", "file_path": "/x.md",
+        })
+        pending = tmp_cache.get_joplin_pending("chatgpt")
+        assert len(pending) == 1
+        assert pending[0][1]["joplin_note_id"] == "note-123"

tests/test_claude_code.py

@@ -0,0 +1,230 @@
+"""Unit tests for the Claude Code session provider."""
+
+import json
+
+import pytest
+
+from src.blocks import (
+    BLOCK_TYPE_COLLAPSED,
+    BLOCK_TYPE_TEXT,
+    BLOCK_TYPE_THINKING,
+    BLOCK_TYPE_TOOL_RESULT,
+    BLOCK_TYPE_TOOL_USE,
+    render_blocks_to_markdown,
+)
+from src.loss_report import LossReport
+from src.providers.claude_code import ClaudeCodeProvider
+
+
+def _write_session(tmp_path, records, project_dir="-home-jesse-myproj", name="abc-123"):
+    proj = tmp_path / project_dir
+    proj.mkdir(parents=True, exist_ok=True)
+    f = proj / f"{name}.jsonl"
+    f.write_text("\n".join(json.dumps(r) for r in records), encoding="utf-8")
+    return f
+
+
+def _records():
+    """A representative session: title, harness noise, dialogue, tool traffic."""
+    return [
+        {"type": "ai-title", "aiTitle": "First title", "sessionId": "abc-123"},
+        {
+            "type": "user",
+            "cwd": "/home/jesse/myproj",
+            "timestamp": "2026-05-01T10:00:00.000Z",
+            "message": {
+                "role": "user",
+                "content": (
+                    "<local-command-caveat>Caveat: ignore</local-command-caveat>"
+                    "<command-name>/model</command-name>"
+                    "<local-command-stdout>Set model</local-command-stdout>"
+                    "Please review my project."
+                ),
+            },
+        },
+        {
+            "type": "assistant",
+            "timestamp": "2026-05-01T10:00:05.000Z",
+            "message": {
+                "role": "assistant",
+                "content": [
+                    {"type": "thinking", "thinking": "Let me think about this."},
+                    {"type": "text", "text": "I'll review it now."},
+                    {"type": "tool_use", "id": "t1", "name": "Read", "input": {"file_path": "/x"}},
+                    {"type": "tool_use", "id": "t2", "name": "Read", "input": {"file_path": "/y"}},
+                    {"type": "tool_use", "id": "t3", "name": "Bash", "input": {"command": "ls"}},
+                ],
+            },
+        },
+        {
+            "type": "user",
+            "timestamp": "2026-05-01T10:00:08.000Z",
+            "message": {
+                "role": "user",
+                "content": [
+                    {"type": "tool_result", "tool_use_id": "t1", "content": "file contents " * 50},
+                ],
+            },
+        },
+        {
+            "type": "assistant",
+            "timestamp": "2026-05-01T10:00:20.000Z",
+            "message": {
+                "role": "assistant",
+                "content": [{"type": "text", "text": "Here is my review: all good."}],
+            },
+        },
+        # Harness records — skipped
+        {"type": "file-history-snapshot", "snapshot": {"big": "blob"}},
+        {"type": "permission-mode", "mode": "plan"},
+        # Subagent and meta records — skipped
+        {
+            "type": "assistant",
+            "isSidechain": True,
+            "message": {"role": "assistant", "content": [{"type": "text", "text": "subagent noise"}]},
+        },
+        {
+            "type": "user",
+            "isMeta": True,
+            "message": {"role": "user", "content": "meta noise"},
+        },
+        # Later title wins
+        {"type": "ai-title", "aiTitle": "Review my project", "sessionId": "abc-123"},
+    ]
+
+
+class TestClaudeCodeProvider:
+    def _provider(self, tmp_path, policy="placeholder"):
+        return ClaudeCodeProvider(projects_dir=tmp_path, hidden_content=policy)
+
+    def test_listing_metadata(self, tmp_path):
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path)
+        convs = p.fetch_all_conversations()
+        assert len(convs) == 1
+        c = convs[0]
+        assert c["id"] == "abc-123"
+        assert c["title"] == "Review my project"  # last ai-title wins
+        assert c["_project_name"] == "myproj"  # from cwd basename
+        assert c["created_at"] == "2026-05-01T10:00:00.000Z"
+        assert c["updated_at"]  # file mtime
+
+    def test_empty_files_skipped(self, tmp_path):
+        proj = tmp_path / "-home-x"
+        proj.mkdir()
+        (proj / "empty.jsonl").write_text("")
+        p = self._provider(tmp_path)
+        assert p.fetch_all_conversations() == []
+
+    def test_normalize_prose_only_default(self, tmp_path):
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path)
+        p.fetch_all_conversations()
+        raw = p.get_conversation("abc-123")
+        result = p.normalize_conversation(raw)
+
+        assert result["provider"] == "claude-code"
+        assert result["title"] == "Review my project"
+        assert result["project"] == "myproj"
+
+        roles = [m["role"] for m in result["messages"]]
+        assert roles == ["user", "assistant", "assistant"]
+
+        # Harness tags stripped, dialogue kept
+        user_text = result["messages"][0]["blocks"][0]["text"]
+        assert user_text == "Please review my project."
+        assert "Caveat" not in user_text
+
+        # Tool activity collapsed into one grouped placeholder on the
+        # assistant message that ran the tools (includes the tool_result
+        # bytes from the following user record).
+        first_assistant = result["messages"][1]
+        types = [b["type"] for b in first_assistant["blocks"]]
+        assert types == [BLOCK_TYPE_TEXT, BLOCK_TYPE_COLLAPSED]
+        collapsed = first_assistant["blocks"][1]
+        assert "3 calls" in collapsed["origin"]
+        assert "Read ×2" in collapsed["origin"]
+        assert "Bash ×1" in collapsed["origin"]
+        assert collapsed["size_bytes"] > 500
+
+        # Thinking dropped without placeholder; sidechain/meta absent
+        all_blocks = [b for m in result["messages"] for b in m["blocks"]]
+        assert not any(b["type"] == BLOCK_TYPE_THINKING for b in all_blocks)
+        assert not any(
+            "subagent noise" in (b.get("text") or "") or "meta noise" in (b.get("text") or "")
+            for b in all_blocks
+        )
+
+    def test_collapsed_counted_in_loss_report(self, tmp_path):
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path)
+        p.fetch_all_conversations()
+        report = LossReport()
+        p.normalize_conversation(p.get_conversation("abc-123"), report)
+        assert report.collapsed["Read"] == 2
+        assert report.collapsed["Bash"] == 1
+        assert report.collapsed["tool_result"] == 1
+        assert report.collapsed["thinking"] == 1
+        assert report.collapsed_bytes > 500
+
+    def test_full_policy_keeps_everything(self, tmp_path):
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path, policy="full")
+        p.fetch_all_conversations()
+        result = p.normalize_conversation(p.get_conversation("abc-123"))
+        all_blocks = [b for m in result["messages"] for b in m["blocks"]]
+        types = {b["type"] for b in all_blocks}
+        assert BLOCK_TYPE_THINKING in types
+        assert BLOCK_TYPE_TOOL_USE in types
+        assert BLOCK_TYPE_TOOL_RESULT in types
+        assert BLOCK_TYPE_COLLAPSED not in types
+
+    def test_updated_at_consistent_with_listing(self, tmp_path):
+        """Listing and normalized updated_at must match or the cache would
+        consider every session stale on every run (mtime vs last timestamp)."""
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path)
+        listing = p.fetch_all_conversations()[0]
+        normalized = p.normalize_conversation(p.get_conversation("abc-123"))
+        assert normalized["updated_at"] == listing["updated_at"]
+
+    def test_title_falls_back_to_first_prompt(self, tmp_path):
+        records = [r for r in _records() if r.get("type") != "ai-title"]
+        _write_session(tmp_path, records)
+        p = self._provider(tmp_path)
+        p.fetch_all_conversations()
+        result = p.normalize_conversation(p.get_conversation("abc-123"))
+        assert result["title"] == "Please review my project."
+
+    def test_unknown_block_type_surfaces(self, tmp_path):
+        records = [
+            {
+                "type": "assistant",
+                "timestamp": "2026-05-01T10:00:00.000Z",
+                "cwd": "/home/jesse/myproj",
+                "message": {
+                    "role": "assistant",
+                    "content": [
+                        {"type": "text", "text": "hello"},
+                        {"type": "future_block_xyz", "data": 1},
+                    ],
+                },
+            },
+        ]
+        _write_session(tmp_path, records)
+        p = self._provider(tmp_path)
+        p.fetch_all_conversations()
+        report = LossReport()
+        result = p.normalize_conversation(p.get_conversation("abc-123"), report)
+        assert report.unknown_blocks["claude-code.future_block_xyz"] == 1
+        rendered = render_blocks_to_markdown(result["messages"][0]["blocks"])
+        assert "Unsupported content" in rendered
+
+    def test_collapsed_placeholder_renders_grouped_line(self, tmp_path):
+        _write_session(tmp_path, _records())
+        p = self._provider(tmp_path)
+        p.fetch_all_conversations()
+        result = p.normalize_conversation(p.get_conversation("abc-123"))
+        rendered = render_blocks_to_markdown(result["messages"][1]["blocks"])
+        assert "> 🔧 **Tool output** — `3 calls: Read ×2, Bash ×1`" in rendered
+        assert "omitted (EXPORTER_HIDDEN_CONTENT=full to keep)" in rendered

tests/test_cli.py

@@ -124,6 +124,176 @@ class TestExportSinceValidation:
                 "CHATGPT_SESSION_TOKEN": "eyJtesttoken",
                 "CACHE_DIR": str(tmp_path),
                 "EXPORT_DIR": str(tmp_path / "exports"),
+                # This test hits the real (failing) auth endpoint with retries;
+                # don't add politeness pacing on top of the backoff sleeps.
+                "REQUEST_DELAY": "0",
             },
         )
         assert "Invalid --since date" not in result.output
+
+    def test_max_conversations_zero_rejected(self, tmp_path):
+        """--max-conversations uses IntRange(min=1); 0 must be rejected by click."""
+        self._pre_populated_cache(tmp_path)
+
+        runner = CliRunner(mix_stderr=True)
+        result = runner.invoke(
+            cli,
+            ["--no-log-file", "export", "--max-conversations", "0"],
+            env={
+                "CHATGPT_SESSION_TOKEN": "eyJtesttoken",
+                "CACHE_DIR": str(tmp_path),
+                "EXPORT_DIR": str(tmp_path / "exports"),
+            },
+        )
+        assert result.exit_code == 2
+        assert "max-conversations" in result.output
+
+
+# ---------------------------------------------------------------------------
+# LossReport summary
+# ---------------------------------------------------------------------------
+
+
+class TestLossReportSummary:
+    """The LossReport's format_summary() pinned format covers zero, top-5, and overflow cases."""
+
+    def test_zero_summary_uses_none_sentinel(self):
+        from src.loss_report import LossReport
+
+        report = LossReport()
+        out = report.format_summary()
+        assert "[export] Run summary:" in out
+        assert "conversations:        0" in out
+        assert "messages rendered:    0" in out
+        # All three "(none)" sentinels present — never empty parens
+        # (unknown blocks, extraction failures, collapsed by policy)
+        assert out.count("(none)") == 3
+
+    def test_top_5_breakdown(self):
+        from src.loss_report import LossReport
+
+        report = LossReport()
+        for raw_type in ("a", "b", "c", "d", "e", "f", "g"):
+            report.record_unknown(raw_type)
+            if raw_type == "a":
+                # Make 'a' the most common
+                for _ in range(4):
+                    report.record_unknown("a")
+        out = report.format_summary()
+        # Top entry shown
+        assert "a=5" in out
+        # Overflow line present (7 types, top 5 + 2 more)
+        assert "+ 2 more types" in out
+
+    def test_messages_and_conversations_recorded(self):
+        from src.loss_report import LossReport
+
+        report = LossReport()
+        report.record_conversation()
+        report.record_message()
+        report.record_message()
+        out = report.format_summary()
+        assert "conversations:        1" in out
+        assert "messages rendered:    2" in out
+
+
+# ---------------------------------------------------------------------------
+# prune command
+# ---------------------------------------------------------------------------
+
+
+class TestPrune:
+    """prune deletes export files not referenced by the manifest."""
+
+    def _setup(self, tmp_path):
+        """Cache with one referenced file; one stale file + empty-dir candidate."""
+        cache = Cache(tmp_path / "cache")
+        cache.acknowledge_tos()
+        export_dir = tmp_path / "exports"
+        keep = export_dir / "chatgpt" / "proj.2026" / "keep.md"
+        keep.parent.mkdir(parents=True)
+        keep.write_text("kept")
+        stale = export_dir / "chatgpt" / "proj" / "2026" / "old-layout.md"
+        stale.parent.mkdir(parents=True)
+        stale.write_text("stale")
+        cache.mark_exported("chatgpt", "conv-1", {"file_path": str(keep)})
+        return export_dir, keep, stale
+
+    def _invoke(self, tmp_path, *args):
+        runner = CliRunner(mix_stderr=True)
+        return runner.invoke(
+            cli,
+            ["--no-log-file", "prune", *args],
+            env={
+                "CACHE_DIR": str(tmp_path / "cache"),
+                "EXPORT_DIR": str(tmp_path / "exports"),
+            },
+        )
+
+    def test_dry_run_lists_but_keeps_files(self, tmp_path):
+        export_dir, keep, stale = self._setup(tmp_path)
+        result = self._invoke(tmp_path, "--dry-run")
+        assert result.exit_code == 0
+        assert "old-layout.md" in result.output
+        assert "Dry run" in result.output
+        assert stale.exists() and keep.exists()
+
+    def test_yes_deletes_stale_keeps_referenced_sweeps_dirs(self, tmp_path):
+        export_dir, keep, stale = self._setup(tmp_path)
+        result = self._invoke(tmp_path, "--yes")
+        assert result.exit_code == 0
+        assert not stale.exists()
+        assert keep.exists()
+        # Old-layout dirs are now empty and swept
+        assert not (export_dir / "chatgpt" / "proj").exists()
+
+    def test_refuses_with_empty_manifest(self, tmp_path):
+        """Footgun guard: after cache --clear, prune must not wipe the archive."""
+        cache = Cache(tmp_path / "cache")
+        cache.acknowledge_tos()
+        export_dir = tmp_path / "exports"
+        f = export_dir / "chatgpt" / "a.md"
+        f.parent.mkdir(parents=True)
+        f.write_text("data")
+        result = self._invoke(tmp_path, "--yes")
+        assert result.exit_code == 1
+        assert "Refusing to prune" in result.output
+        assert f.exists()
+
+    def test_aborts_without_confirmation(self, tmp_path):
+        export_dir, keep, stale = self._setup(tmp_path)
+        runner = CliRunner(mix_stderr=True)
+        result = runner.invoke(
+            cli,
+            ["--no-log-file", "prune"],
+            input="n\n",
+            env={
+                "CACHE_DIR": str(tmp_path / "cache"),
+                "EXPORT_DIR": str(tmp_path / "exports"),
+            },
+        )
+        assert result.exit_code == 0
+        assert "Aborted" in result.output
+        assert stale.exists()
+
+
+class TestCanaryCommand:
+    """`canary` wiring — no live API calls (no tokens configured)."""
+
+    def test_no_tokens_exits_nonzero(self, tmp_path):
+        cache = Cache(tmp_path / "cache")
+        cache.acknowledge_tos()
+        runner = CliRunner(mix_stderr=True)
+        result = runner.invoke(
+            cli,
+            ["--no-log-file", "canary"],
+            env={
+                "CACHE_DIR": str(tmp_path / "cache"),
+                "EXPORT_DIR": str(tmp_path / "exports"),
+                # Empty so .env (override=False) can't repopulate real tokens.
+                "CHATGPT_SESSION_TOKEN": "",
+                "CLAUDE_SESSION_KEY": "",
+            },
+        )
+        assert result.exit_code == 1
+        assert "No web-API provider tokens" in result.output

tests/test_config.py

@@ -54,3 +54,65 @@ class TestValidateChatGPTToken:
             result = _validate_chatgpt_token("notajwttoken")
         assert any("does not look like a JWT" in r.message for r in caplog.records)
         assert result is None
+
+
+class TestSessionLimiterConfig:
+    """MAX_CONVERSATIONS_PER_RUN and REQUEST_DELAY parsing in load_config."""
+
+    def _load(self, monkeypatch, tmp_path, **env):
+        from src.config import load_config
+        monkeypatch.setenv("EXPORT_DIR", str(tmp_path / "exports"))
+        monkeypatch.setenv("CACHE_DIR", str(tmp_path / "cache"))
+        for key in (
+            "MAX_CONVERSATIONS_PER_RUN",
+            "REQUEST_DELAY",
+            "EXPORTER_HIDDEN_CONTENT",
+            "EXPORTER_DOWNLOAD_MEDIA",
+        ):
+            monkeypatch.delenv(key, raising=False)
+        for key, value in env.items():
+            monkeypatch.setenv(key, value)
+        return load_config()
+
+    def test_defaults(self, monkeypatch, tmp_path):
+        cfg = self._load(monkeypatch, tmp_path)
+        assert cfg.max_conversations is None
+        assert cfg.request_delay == 1.0
+        assert cfg.hidden_content == "placeholder"
+        assert cfg.download_media == "images"
+
+    def test_download_media_valid(self, monkeypatch, tmp_path):
+        cfg = self._load(monkeypatch, tmp_path, EXPORTER_DOWNLOAD_MEDIA="all")
+        assert cfg.download_media == "all"
+
+    def test_download_media_invalid_raises(self, monkeypatch, tmp_path):
+        from src.config import ConfigError
+        with pytest.raises(ConfigError, match="EXPORTER_DOWNLOAD_MEDIA"):
+            self._load(monkeypatch, tmp_path, EXPORTER_DOWNLOAD_MEDIA="sometimes")
+
+    def test_valid_values(self, monkeypatch, tmp_path):
+        cfg = self._load(
+            monkeypatch, tmp_path,
+            MAX_CONVERSATIONS_PER_RUN="25", REQUEST_DELAY="0.5",
+        )
+        assert cfg.max_conversations == 25
+        assert cfg.request_delay == 0.5
+
+    def test_zero_delay_allowed(self, monkeypatch, tmp_path):
+        cfg = self._load(monkeypatch, tmp_path, REQUEST_DELAY="0")
+        assert cfg.request_delay == 0.0
+
+    def test_non_integer_cap_raises(self, monkeypatch, tmp_path):
+        from src.config import ConfigError
+        with pytest.raises(ConfigError, match="MAX_CONVERSATIONS_PER_RUN"):
+            self._load(monkeypatch, tmp_path, MAX_CONVERSATIONS_PER_RUN="lots")
+
+    def test_zero_cap_raises(self, monkeypatch, tmp_path):
+        from src.config import ConfigError
+        with pytest.raises(ConfigError, match="at least 1"):
+            self._load(monkeypatch, tmp_path, MAX_CONVERSATIONS_PER_RUN="0")
+
+    def test_negative_delay_raises(self, monkeypatch, tmp_path):
+        from src.config import ConfigError
+        with pytest.raises(ConfigError, match="REQUEST_DELAY"):
+            self._load(monkeypatch, tmp_path, REQUEST_DELAY="-1")

tests/test_exporters.py

@@ -1,4 +1,4 @@
-"""Unit tests for src/exporters/."""
+"""Unit tests for src/exporters/ and src/blocks.py."""
 
 import json
 import os
@@ -7,6 +7,23 @@ from pathlib import Path
 
 import pytest
 
+from src.blocks import (
+    BLOCK_TYPE_TEXT,
+    UNKNOWN_REASON_EXTRACTION_FAILED,
+    UNKNOWN_REASON_UNKNOWN_TYPE,
+    _blockquote_prefix,
+    _safe_fence,
+    make_code_block,
+    make_file_placeholder,
+    make_hidden_context_marker,
+    make_image_placeholder,
+    make_text_block,
+    make_thinking_block,
+    make_tool_result_block,
+    make_tool_use_block,
+    make_unknown_block,
+    render_blocks_to_markdown,
+)
 from src.exporters.markdown import MarkdownExporter, _yaml_escape, _format_timestamp
 from src.exporters.json_export import JSONExporter
 
@@ -122,7 +139,7 @@ class TestMarkdownFilenameGeneration:
     def test_year_in_path(self, tmp_path):
         exp = MarkdownExporter(tmp_path)
         path = exp.export(SAMPLE_CONV)
-        assert "/2024/" in str(path)
+        assert ".2024/" in str(path)
 
     def test_output_structure_provider_project(self, tmp_path):
         exp = MarkdownExporter(tmp_path, output_structure="provider/project")
@@ -250,3 +267,271 @@ class TestFormatTimestamp:
 
     def test_empty_string(self):
         assert _format_timestamp("") == ""
+
+
+# ---------------------------------------------------------------------------
+# Block helpers and rendering
+# ---------------------------------------------------------------------------
+
+
+class TestSafeFence:
+    def test_minimum_three_backticks(self):
+        assert _safe_fence("plain text") == "```"
+
+    def test_four_backticks_when_three_in_content(self):
+        assert _safe_fence("here ``` is a fence") == "````"
+
+    def test_five_backticks_when_four_in_content(self):
+        assert _safe_fence("here ```` is four") == "`````"
+
+    def test_handles_empty_string(self):
+        assert _safe_fence("") == "```"
+
+    def test_handles_run_at_end(self):
+        # Trailing run still counted
+        assert _safe_fence("text ending in ```") == "````"
+
+
+class TestBlockquotePrefix:
+    def test_single_line(self):
+        assert _blockquote_prefix("hello") == "> hello"
+
+    def test_multi_line(self):
+        assert _blockquote_prefix("a\nb\nc") == "> a\n> b\n> c"
+
+    def test_empty_lines_become_naked_quote_marker(self):
+        assert _blockquote_prefix("a\n\nb") == "> a\n>\n> b"
+
+    def test_empty_string(self):
+        assert _blockquote_prefix("") == ">"
+
+
+class TestBlockConstructors:
+    def test_make_text_block_returns_none_for_empty(self):
+        assert make_text_block("") is None
+        assert make_text_block("   ") is None
+
+    def test_make_text_block_returns_dict(self):
+        b = make_text_block("hello")
+        assert b == {"type": "text", "text": "hello"}
+
+    def test_make_code_block_returns_none_for_empty(self):
+        assert make_code_block("") is None
+
+    def test_make_thinking_block_returns_none_for_empty(self):
+        assert make_thinking_block("") is None
+
+
+class TestRenderBlocks:
+    def test_text_block_renders_as_paragraph(self):
+        out = render_blocks_to_markdown([make_text_block("Hello world")])
+        assert out == "Hello world"
+
+    def test_blocks_separated_by_blank_line(self):
+        out = render_blocks_to_markdown(
+            [make_text_block("first"), make_text_block("second")]
+        )
+        assert out == "first\n\nsecond"
+
+    def test_code_block_with_language(self):
+        out = render_blocks_to_markdown([make_code_block("print(1)", language="python")])
+        assert "```python" in out
+        assert "print(1)" in out
+
+    def test_thinking_block_uses_blockquote(self):
+        out = render_blocks_to_markdown([make_thinking_block("step 1\nstep 2")])
+        assert "**💭 Reasoning**" in out
+        assert "> step 1" in out
+        assert "> step 2" in out
+
+    def test_tool_use_renders_as_blockquote_with_safe_fence(self):
+        out = render_blocks_to_markdown(
+            [make_tool_use_block("search", {"query": "test"})]
+        )
+        assert "> 🔧 **Tool: search**" in out
+        # Every line of the body is blockquote-prefixed
+        assert "> ```json" in out
+        assert "> }" in out
+
+    def test_tool_use_with_multiline_input(self):
+        out = render_blocks_to_markdown(
+            [make_tool_use_block("complex", {"a": 1, "b": [{"x": "y"}]})]
+        )
+        # Prefix every line of multi-line JSON
+        for line in out.split("\n"):
+            assert line.startswith(">") or line == ""
+
+    def test_tool_result_success_uses_outbox_icon(self):
+        out = render_blocks_to_markdown([make_tool_result_block("OK")])
+        assert "📤 **Result**" in out
+        assert "❌" not in out
+
+    def test_tool_result_error_uses_x_icon(self):
+        out = render_blocks_to_markdown([make_tool_result_block("oops", is_error=True)])
+        assert "❌ **Result (error)**" in out
+        assert "📤" not in out
+
+    def test_tool_result_with_tool_name_in_header(self):
+        out = render_blocks_to_markdown(
+            [make_tool_result_block("done", tool_name="container.exec")]
+        )
+        assert "📤 **Result: container.exec**" in out
+
+    def test_tool_result_error_with_tool_name(self):
+        out = render_blocks_to_markdown(
+            [make_tool_result_block("503", tool_name="web", is_error=True)]
+        )
+        assert "❌ **Result (error): web**" in out
+
+    def test_tool_result_summary_renders_as_italic_line(self):
+        out = render_blocks_to_markdown(
+            [
+                make_tool_result_block(
+                    "output",
+                    tool_name="container.exec",
+                    summary="Reading skill documentation",
+                )
+            ]
+        )
+        # Summary line is italic, lives between header and fence,
+        # all inside the blockquote prefix.
+        assert "> *Reading skill documentation*" in out
+        # Order: header before summary before fence
+        header_idx = out.index("Result: container.exec")
+        summary_idx = out.index("Reading skill documentation")
+        fence_idx = out.index("output")
+        assert header_idx < summary_idx < fence_idx
+
+    def test_image_placeholder_rendering(self):
+        out = render_blocks_to_markdown(
+            [make_image_placeholder(ref="file-123", source="user_upload")]
+        )
+        assert "🖼️ **Image attached**" in out
+        assert "`file-123`" in out
+        assert "user_upload" in out
+        assert "content not preserved" in out
+
+    def test_file_placeholder_with_metadata(self):
+        out = render_blocks_to_markdown(
+            [make_file_placeholder(ref="sediment://x", mime="audio/wav", size_bytes=10240, duration_seconds=2.5)]
+        )
+        assert "📎 **File attached**" in out
+        assert "audio/wav" in out
+        assert "KB" in out
+        assert "2.50s" in out
+
+    def test_unknown_block_renders_with_keys(self):
+        out = render_blocks_to_markdown(
+            [
+                make_unknown_block(
+                    raw_type="future_x",
+                    observed_keys=["foo", "bar"],
+                    reason=UNKNOWN_REASON_UNKNOWN_TYPE,
+                )
+            ]
+        )
+        assert "⚠️ **Unsupported content**" in out
+        assert "future_x" in out
+        assert "`foo`" in out
+        assert "`bar`" in out
+
+    def test_unknown_extraction_failed_includes_summary(self):
+        out = render_blocks_to_markdown(
+            [
+                make_unknown_block(
+                    raw_type="audio_transcription",
+                    observed_keys=["asset_pointer"],
+                    reason=UNKNOWN_REASON_EXTRACTION_FAILED,
+                    summary="expected key 'text' not found",
+                )
+            ]
+        )
+        assert "extraction_failed" in out
+        assert "expected key 'text' not found" in out
+
+    def test_hidden_context_marker(self):
+        out = render_blocks_to_markdown(
+            [make_hidden_context_marker("user_editable_context")]
+        )
+        assert "ℹ️ **Hidden context**" in out
+        assert "`user_editable_context`" in out
+
+    def test_safe_fence_prevents_runaway_code_block(self):
+        # Content contains an unbalanced opening fence — without _safe_fence
+        # this would corrupt downstream rendering.
+        evil_content = "before\n```Follow\ntext\nraw is: \"```"
+        block = make_code_block(evil_content)
+        out = render_blocks_to_markdown([block, make_text_block("after")])
+        # The 4-backtick wrap should be present
+        assert "````" in out
+        # The "after" text should appear OUTSIDE any code block — it follows
+        # the closing ```` fence.
+        assert out.endswith("after")
+
+    def test_block_order_preserved(self):
+        blocks = [
+            make_text_block("a"),
+            make_image_placeholder(ref="r1", source="user_upload"),
+            make_text_block("b"),
+        ]
+        out = render_blocks_to_markdown(blocks)
+        assert out.index("a") < out.index("Image attached")
+        assert out.index("Image attached") < out.index("b")
+
+
+# ---------------------------------------------------------------------------
+# Markdown exporter with blocks
+# ---------------------------------------------------------------------------
+
+
+SAMPLE_CONV_BLOCKS = {
+    "id": "blocks12345",
+    "title": "Blocks Conversation",
+    "provider": "claude",
+    "project": None,
+    "created_at": "2024-06-10T14:32:00Z",
+    "updated_at": "2024-06-10T15:00:00Z",
+    "message_count": 1,
+    "messages": [
+        {
+            "role": "assistant",
+            "content_type": "text",
+            "timestamp": None,
+            "blocks": [
+                {"type": "text", "text": "Here is the answer."},
+                {"type": "tool_use", "name": "search", "input": {"q": "x"}, "tool_id": "t1"},
+            ],
+        }
+    ],
+}
+
+
+class TestMarkdownExporterWithBlocks:
+    def test_renders_blocks(self, tmp_path):
+        exp = MarkdownExporter(tmp_path)
+        path = exp.export(SAMPLE_CONV_BLOCKS)
+        body = path.read_text()
+        assert "Here is the answer." in body
+        assert "🔧 **Tool: search**" in body
+
+    def test_falls_back_to_content_when_blocks_missing(self, tmp_path):
+        # Backward-compat: messages with `content` only (no `blocks`) still render.
+        exp = MarkdownExporter(tmp_path)
+        path = exp.export(SAMPLE_CONV)  # SAMPLE_CONV has content only, no blocks
+        body = path.read_text()
+        assert "Hello, how are you?" in body
+
+    def test_skips_messages_with_neither_blocks_nor_content(self, tmp_path):
+        conv = {
+            **SAMPLE_CONV_BLOCKS,
+            "messages": [
+                {"role": "user", "content_type": "text", "timestamp": None, "blocks": []},
+                {"role": "assistant", "content_type": "text", "timestamp": None, "blocks": [
+                    {"type": "text", "text": "I am here."}
+                ]},
+            ],
+        }
+        exp = MarkdownExporter(tmp_path)
+        path = exp.export(conv)
+        body = path.read_text()
+        assert "I am here." in body

tests/test_joplin.py

@@ -5,7 +5,7 @@ from unittest.mock import MagicMock, patch
 import pytest
 import requests
 
-from src.joplin import JoplinClient, JoplinError, _http_error_message, _timeout_message, notebook_title
+from src.joplin import JoplinClient, JoplinError, _http_error_message, _timeout_message, notebook_path
 
 
 # ---------------------------------------------------------------------------
@@ -31,25 +31,29 @@ def _mock_response(json_data=None, text="", status_code=200):
 
 
 # ---------------------------------------------------------------------------
-# notebook_title helper
+# notebook_path helper
 # ---------------------------------------------------------------------------
 
 
-class TestNotebookTitle:
+class TestNotebookPath:
     def test_no_project(self):
-        assert notebook_title("chatgpt", None) == "ChatGPT - No Project"
+        assert notebook_path("chatgpt", None) == ("AI-ChatGPT", "No Project")
 
     def test_no_project_string(self):
-        assert notebook_title("chatgpt", "no-project") == "ChatGPT - No Project"
+        assert notebook_path("chatgpt", "no-project") == ("AI-ChatGPT", "No Project")
 
     def test_project_with_hyphens(self):
-        assert notebook_title("chatgpt", "my-project") == "ChatGPT - My Project"
+        assert notebook_path("chatgpt", "my-project") == ("AI-ChatGPT", "My Project")
 
     def test_claude_provider(self):
-        assert notebook_title("claude", "budget-tracker") == "Claude - Budget Tracker"
+        assert notebook_path("claude", "budget-tracker") == ("AI-Claude", "Budget Tracker")
 
     def test_multi_word_project(self):
-        assert notebook_title("claude", "ai-research-notes") == "Claude - Ai Research Notes"
+        assert notebook_path("claude", "ai-research-notes") == ("AI-Claude", "Ai Research Notes")
+
+    def test_returns_tuple(self):
+        result = notebook_path("chatgpt", "some-project")
+        assert isinstance(result, tuple) and len(result) == 2
 
 
 # ---------------------------------------------------------------------------
@@ -236,18 +240,30 @@ class TestListNotebooks:
 
 
 class TestGetOrCreateNotebook:
-    def test_returns_existing_notebook_id(self):
+    def test_returns_existing_root_notebook_id(self):
         client = _make_client()
         with patch("requests.get") as mock_get:
             mock_get.return_value = _mock_response(
                 json_data={
-                    "items": [{"id": "nb-existing", "title": "ChatGPT - No Project"}],
+                    "items": [{"id": "nb-existing", "title": "AI-ChatGPT", "parent_id": ""}],
                     "has_more": False,
                 }
             )
-            nb_id = client.get_or_create_notebook("ChatGPT - No Project")
+            nb_id = client.get_or_create_notebook("AI-ChatGPT")
         assert nb_id == "nb-existing"
 
+    def test_returns_existing_child_notebook_id(self):
+        client = _make_client()
+        with patch("requests.get") as mock_get:
+            mock_get.return_value = _mock_response(
+                json_data={
+                    "items": [{"id": "nb-child", "title": "No Project", "parent_id": "nb-parent"}],
+                    "has_more": False,
+                }
+            )
+            nb_id = client.get_or_create_notebook("No Project", parent_id="nb-parent")
+        assert nb_id == "nb-child"
+
     def test_creates_new_notebook_when_not_found(self):
         client = _make_client()
         with patch("requests.get") as mock_get, patch("requests.post") as mock_post:
@@ -255,26 +271,103 @@ class TestGetOrCreateNotebook:
                 json_data={"items": [], "has_more": False}
             )
             mock_post.return_value = _mock_response(
-                json_data={"id": "nb-new", "title": "ChatGPT - New Project"}
+                json_data={"id": "nb-new", "title": "AI-ChatGPT"}
             )
-            nb_id = client.get_or_create_notebook("ChatGPT - New Project")
+            nb_id = client.get_or_create_notebook("AI-ChatGPT")
         assert nb_id == "nb-new"
         mock_post.assert_called_once()
 
+    def test_creates_child_notebook_with_parent_id(self):
+        client = _make_client()
+        with patch("requests.get") as mock_get, patch("requests.post") as mock_post:
+            mock_get.return_value = _mock_response(
+                json_data={"items": [], "has_more": False}
+            )
+            mock_post.return_value = _mock_response(
+                json_data={"id": "nb-child", "title": "My Project"}
+            )
+            nb_id = client.get_or_create_notebook("My Project", parent_id="nb-parent")
+        assert nb_id == "nb-child"
+        _, kwargs = mock_post.call_args
+        assert kwargs["json"]["parent_id"] == "nb-parent"
+
+    def test_does_not_include_parent_id_for_root(self):
+        client = _make_client()
+        with patch("requests.get") as mock_get, patch("requests.post") as mock_post:
+            mock_get.return_value = _mock_response(json_data={"items": [], "has_more": False})
+            mock_post.return_value = _mock_response(json_data={"id": "nb-root", "title": "AI-Claude"})
+            client.get_or_create_notebook("AI-Claude")
+        _, kwargs = mock_post.call_args
+        assert "parent_id" not in kwargs["json"]
+
     def test_caches_notebook_after_first_load(self):
         client = _make_client()
         with patch("requests.get") as mock_get:
             mock_get.return_value = _mock_response(
                 json_data={
-                    "items": [{"id": "nb1", "title": "Claude - No Project"}],
+                    "items": [{"id": "nb1", "title": "AI-Claude", "parent_id": ""}],
                     "has_more": False,
                 }
             )
             # Call twice — GET /folders should only happen once
-            client.get_or_create_notebook("Claude - No Project")
+            client.get_or_create_notebook("AI-Claude")
-            client.get_or_create_notebook("Claude - No Project")
+            client.get_or_create_notebook("AI-Claude")
         assert mock_get.call_count == 1
 
+    def test_different_parent_ids_are_distinct_cache_entries(self):
+        """Same title under different parents are different notebooks."""
+        client = _make_client()
+        with patch("requests.get") as mock_get:
+            mock_get.return_value = _mock_response(
+                json_data={
+                    "items": [
+                        {"id": "nb-a", "title": "No Project", "parent_id": "parent-chatgpt"},
+                        {"id": "nb-b", "title": "No Project", "parent_id": "parent-claude"},
+                    ],
+                    "has_more": False,
+                }
+            )
+            id_a = client.get_or_create_notebook("No Project", parent_id="parent-chatgpt")
+            id_b = client.get_or_create_notebook("No Project", parent_id="parent-claude")
+        assert id_a == "nb-a"
+        assert id_b == "nb-b"
+
+
+class TestGetOrCreateNotebookPath:
+    def test_creates_two_level_path(self):
+        client = _make_client()
+        with patch("requests.get") as mock_get, patch("requests.post") as mock_post:
+            mock_get.return_value = _mock_response(json_data={"items": [], "has_more": False})
+            mock_post.side_effect = [
+                _mock_response(json_data={"id": "nb-parent", "title": "AI-ChatGPT"}),
+                _mock_response(json_data={"id": "nb-child", "title": "No Project"}),
+            ]
+            leaf_id = client.get_or_create_notebook_path(["AI-ChatGPT", "No Project"])
+        assert leaf_id == "nb-child"
+        assert mock_post.call_count == 2
+        # Second POST should use the parent's ID
+        _, kwargs = mock_post.call_args_list[1]
+        assert kwargs["json"]["parent_id"] == "nb-parent"
+
+    def test_reuses_existing_parent_for_new_child(self):
+        client = _make_client()
+        with patch("requests.get") as mock_get, patch("requests.post") as mock_post:
+            mock_get.return_value = _mock_response(
+                json_data={
+                    "items": [{"id": "nb-parent", "title": "AI-Claude", "parent_id": ""}],
+                    "has_more": False,
+                }
+            )
+            mock_post.return_value = _mock_response(
+                json_data={"id": "nb-child", "title": "Budget Tracker"}
+            )
+            leaf_id = client.get_or_create_notebook_path(["AI-Claude", "Budget Tracker"])
+        assert leaf_id == "nb-child"
+        # Only one POST — the parent already existed
+        assert mock_post.call_count == 1
+        _, kwargs = mock_post.call_args
+        assert kwargs["json"]["parent_id"] == "nb-parent"
+
 
 # ---------------------------------------------------------------------------
 # create_note

tests/test_media.py

@@ -0,0 +1,248 @@
+"""Tests for media downloads and Joplin resource rewriting."""
+
+from pathlib import Path
+
+import pytest
+
+from src.blocks import (
+    make_file_placeholder,
+    make_image_placeholder,
+    render_blocks_to_markdown,
+)
+from src.loss_report import LossReport
+from src.media import resolve_media, resolve_media_policy
+from src.providers.base import ProviderError
+from src.providers.chatgpt import parse_asset_file_id
+
+
+# ---------------------------------------------------------------------------
+# Asset reference parsing
+# ---------------------------------------------------------------------------
+
+
+class TestParseAssetFileId:
+    def test_plain_sediment(self):
+        assert parse_asset_file_id("sediment://file_00000000245c71fda5") == "file_00000000245c71fda5"
+
+    def test_generated_image_with_hash_and_page(self):
+        ref = "sediment://8456107fc383a53#file_00000000979c71f685#p_6.png"
+        assert parse_asset_file_id(ref) == "file_00000000979c71f685"
+
+    def test_file_service_scheme(self):
+        assert parse_asset_file_id("file-service://file-AbCdEf") == "file-AbCdEf"
+
+    def test_unrecognised(self):
+        assert parse_asset_file_id("https://example.com/x.png") is None
+        assert parse_asset_file_id("") is None
+        assert parse_asset_file_id(None) is None
+
+
+# ---------------------------------------------------------------------------
+# Media policy
+# ---------------------------------------------------------------------------
+
+
+class TestMediaPolicy:
+    def test_default(self, monkeypatch):
+        monkeypatch.delenv("EXPORTER_DOWNLOAD_MEDIA", raising=False)
+        assert resolve_media_policy() == "images"
+
+    def test_valid(self, monkeypatch):
+        monkeypatch.setenv("EXPORTER_DOWNLOAD_MEDIA", "all")
+        assert resolve_media_policy() == "all"
+
+    def test_invalid_falls_back(self, monkeypatch, caplog):
+        monkeypatch.setenv("EXPORTER_DOWNLOAD_MEDIA", "bogus")
+        assert resolve_media_policy() == "images"
+
+
+# ---------------------------------------------------------------------------
+# resolve_media
+# ---------------------------------------------------------------------------
+
+
+class _FakeProvider:
+    """Minimal provider exposing download_asset + the ref parser."""
+
+    def __init__(self, assets=None, fail_refs=None):
+        self._assets = assets or {}
+        self._fail_refs = fail_refs or {}
+        self.calls = []
+
+    def parse_asset_file_id(self, ref):
+        return parse_asset_file_id(ref)
+
+    def download_asset(self, ref):
+        self.calls.append(ref)
+        if ref in self._fail_refs:
+            raise ProviderError("chatgpt", "download_asset", self._fail_refs[ref])
+        return self._assets[ref]  # (content, mime, file_name)
+
+
+def _conv_with(blocks):
+    return {
+        "id": "conv-1",
+        "title": "Has Media",
+        "provider": "chatgpt",
+        "project": None,
+        "created_at": "2026-05-20T00:00:00+00:00",
+        "messages": [{"role": "user", "blocks": blocks}],
+    }
+
+
+class TestResolveMedia:
+    def test_downloads_image_and_inlines(self, tmp_path):
+        ref = "sediment://file_img1"
+        provider = _FakeProvider({ref: (b"\x89PNG\r\n", "image/png", "x.png")})
+        block = make_image_placeholder(ref=ref, source="user_upload")
+        conv = _conv_with([block])
+        report = LossReport()
+
+        n = resolve_media(conv, provider, tmp_path, "provider/project/year", "images", report)
+
+        assert n == 1
+        assert report.media_downloaded == 1
+        assert block["local_path"] == "media/file_img1.png"
+        rendered = render_blocks_to_markdown([block])
+        assert rendered == "![user_upload](media/file_img1.png)"
+        # File written under the conversation's media/ dir
+        written = list(tmp_path.rglob("media/file_img1.png"))
+        assert written and written[0].read_bytes() == b"\x89PNG\r\n"
+
+    def test_images_policy_skips_files(self, tmp_path):
+        ref = "sediment://file_audio1"
+        provider = _FakeProvider({ref: (b"RIFF", "audio/wav", "a.wav")})
+        block = make_file_placeholder(ref=ref, mime="audio/wav", size_bytes=1000)
+        conv = _conv_with([block])
+        report = LossReport()
+
+        n = resolve_media(conv, provider, tmp_path, "provider/project/year", "images", report)
+        assert n == 0
+        assert "local_path" not in block
+        assert provider.calls == []
+
+    def test_all_policy_downloads_files(self, tmp_path):
+        ref = "sediment://file_audio1"
+        provider = _FakeProvider({ref: (b"RIFFdata", "audio/wav", "a.wav")})
+        block = make_file_placeholder(ref=ref, mime="audio/wav", size_bytes=8)
+        conv = _conv_with([block])
+        report = LossReport()
+
+        n = resolve_media(conv, provider, tmp_path, "provider/project/year", "all", report)
+        assert n == 1
+        assert block["local_path"] == "media/file_audio1.wav"
+        rendered = render_blocks_to_markdown([block])
+        assert "media/file_audio1.wav" in rendered and rendered.startswith("> 📎")
+
+    def test_off_policy_noop(self, tmp_path):
+        provider = _FakeProvider({"sediment://file_x": (b"x", "image/png", None)})
+        block = make_image_placeholder(ref="sediment://file_x", source="user_upload")
+        conv = _conv_with([block])
+        report = LossReport()
+        assert resolve_media(conv, provider, tmp_path, "provider/project/year", "off", report) == 0
+        assert provider.calls == []
+
+    def test_idempotent_uses_disk(self, tmp_path):
+        ref = "sediment://file_img1"
+        provider = _FakeProvider({ref: (b"\x89PNG", "image/png", "x.png")})
+        conv = _conv_with([make_image_placeholder(ref=ref, source="user_upload")])
+        report = LossReport()
+        resolve_media(conv, provider, tmp_path, "provider/project/year", "images", report)
+        assert len(provider.calls) == 1
+
+        # Second run, fresh blocks: file already on disk → no new API call.
+        conv2 = _conv_with([make_image_placeholder(ref=ref, source="user_upload")])
+        resolve_media(conv2, provider, tmp_path, "provider/project/year", "images", LossReport())
+        assert len(provider.calls) == 1  # unchanged
+        assert conv2["messages"][0]["blocks"][0]["local_path"] == "media/file_img1.png"
+
+    def test_failure_keeps_placeholder_and_counts(self, tmp_path):
+        ref = "sediment://file_gone"
+        provider = _FakeProvider(fail_refs={ref: RuntimeError("Signed URL returned HTTP 404")})
+        block = make_image_placeholder(ref=ref, source="model_generated")
+        conv = _conv_with([block])
+        report = LossReport()
+
+        n = resolve_media(conv, provider, tmp_path, "provider/project/year", "images", report)
+        assert n == 0
+        assert "local_path" not in block
+        assert report.media_failed["expired-or-missing"] == 1
+        # Still renders as a placeholder, not a broken image link
+        assert render_blocks_to_markdown([block]).startswith("> 🖼️")
+
+    def test_provider_without_download_asset(self, tmp_path):
+        """claude-code has no remote assets — resolve_media must no-op."""
+        class NoDownload:
+            pass
+        block = make_image_placeholder(ref="sediment://file_x", source="user_upload")
+        conv = _conv_with([block])
+        assert resolve_media(conv, NoDownload(), tmp_path, "provider/project/year", "images", LossReport()) == 0
+
+
+# ---------------------------------------------------------------------------
+# Joplin media rewriting
+# ---------------------------------------------------------------------------
+
+
+class _FakeJoplin:
+    def __init__(self):
+        self.uploaded = []
+        self._n = 0
+
+    def create_resource(self, file_path, title=None):
+        self.uploaded.append(Path(file_path).name)
+        self._n += 1
+        return f"res{self._n}"
+
+
+class TestUploadMediaAndRewrite:
+    def _note(self, tmp_path, body):
+        media = tmp_path / "media"
+        media.mkdir()
+        (media / "file_img1.png").write_bytes(b"\x89PNG")
+        (media / "clip.wav").write_bytes(b"RIFF")
+        return body
+
+    def test_rewrites_image_and_file_links(self, tmp_path):
+        from src.joplin import upload_media_and_rewrite
+        body = self._note(
+            tmp_path,
+            "Look: ![user_upload](media/file_img1.png)\n"
+            "> 📎 **File attached** — [clip.wav](media/clip.wav) (audio/wav)",
+        )
+        client = _FakeJoplin()
+        new_body, res_map = upload_media_and_rewrite(body, tmp_path, client, {})
+
+        assert "![user_upload](:/res1)" in new_body
+        assert "(:/res2)" in new_body
+        assert "media/" not in new_body
+        assert res_map == {"media/file_img1.png": "res1", "media/clip.wav": "res2"}
+        assert len(client.uploaded) == 2
+
+    def test_reuses_known_resource_ids(self, tmp_path):
+        from src.joplin import upload_media_and_rewrite
+        body = self._note(tmp_path, "![x](media/file_img1.png)")
+        client = _FakeJoplin()
+        existing = {"media/file_img1.png": "existing-res"}
+        new_body, res_map = upload_media_and_rewrite(body, tmp_path, client, existing)
+        assert "(:/existing-res)" in new_body
+        assert client.uploaded == []  # no re-upload
+        assert res_map == existing
+
+    def test_missing_file_left_as_link(self, tmp_path):
+        from src.joplin import upload_media_and_rewrite
+        (tmp_path / "media").mkdir()
+        body = "![x](media/gone.png)"
+        client = _FakeJoplin()
+        new_body, res_map = upload_media_and_rewrite(body, tmp_path, client, {})
+        assert new_body == body
+        assert res_map == {}
+        assert client.uploaded == []
+
+    def test_no_media_links_untouched(self, tmp_path):
+        from src.joplin import upload_media_and_rewrite
+        body = "Just text with [a link](https://example.com) and `media/foo` in code."
+        client = _FakeJoplin()
+        new_body, res_map = upload_media_and_rewrite(body, tmp_path, client, {})
+        assert new_body == body
+        assert res_map == {}

tests/test_utils.py

@@ -57,13 +57,13 @@ class TestBuildExportPath:
         path = build_export_path(
             Path("/exports"), "claude", "my-project", "2024-06-01T00:00:00Z", "file.md"
         )
-        assert str(path) == "/exports/claude/my-project/2024/file.md"
+        assert str(path) == "/exports/claude/my-project.2024/file.md"
 
     def test_no_project_uses_no_project_slug(self):
         path = build_export_path(
             Path("/exports"), "chatgpt", None, "2024-06-01T00:00:00Z", "file.md"
         )
-        assert "no-project" in str(path)
+        assert "no-project.2024" in str(path)
 
     def test_provider_project_structure_omits_year(self):
         path = build_export_path(