wg-html-chat

Metadata

Description

## Description
The wg-html-v2 viewer surfaces per-task data but doesn't render chat transcripts (the JSONL conversation history at `~/.wg/chat/<ref>/*.jsonl`). Add a `--chat` flag to opt into rendering chat content, with visibility-aware defaults so a casual `wg html --chat` doesn't accidentally publish the user's conversations.

## Spec

### Flag matrix

| Command | Tasks rendered | Chat transcripts |
|---------|---------------|-------------------|
| `wg html` (default) | all tasks | none — chat task nodes rendered but transcript omitted |
| `wg html --chat` | all tasks | only chats with `visibility = public` |
| `wg html --chat --all` | all tasks | all chats regardless of visibility |
| `wg html --chat --public-only` | only public tasks | only public chats |

The single `visibility` field on the chat task drives transcript inclusion. No separate `chat.share` field — overkill for a single-user codebase.

### Rendering
- Chat task's per-page or side-panel section adds a 'Conversation' block when transcript is included
- Render messages role-by-role with timestamps; code blocks get the same syntax/theme treatment as task descriptions
- Preserve order from the JSONL
- Light sanitization layer: redact obvious secret patterns in transcript text (api-key-shaped strings, absolute paths under `~/.wg/secrets` etc.) with `[redacted]` markers. NOT a security guarantee — just a safety net against accidental shell-history leaks.
- If transcript is OMITTED due to visibility filter, render a single line: 'Chat transcript hidden (visibility: internal). Use --all to include.'

### Footer / header note
Page header should say 'Showing N chat transcripts' when --chat is active. If 0 chats are shown but --chat was passed (because all chats are internal), say so explicitly so the user isn't confused.

## Validation
- [ ] Failing tests written first
- [ ] `wg html` (no --chat) — no transcript content in output, but chat task nodes still appear in the viz
- [ ] `wg html --chat` on a project with mixed-visibility chats — only public chats' transcripts in output; internal chats show 'transcript hidden' message
- [ ] `wg html --chat --all` — all chat transcripts in output
- [ ] `wg html --chat --public-only` — only public tasks AND only public chats
- [ ] Sanitization layer redacts at least: `sk-[a-zA-Z0-9]{20,}` (api-key shape), `OPENAI_API_KEY=...`, paths under `~/.wg/secrets`. Document the regexes; this is best-effort, not a security guarantee.
- [ ] Live smoke: `wg html --chat` on this project, open a few chat task pages in browser, confirm conversation renders with role/timestamp/code-block formatting
- [ ] Counter footer accurate (N chat transcripts shown / M omitted by visibility)
- [ ] cargo build + cargo test pass
- [ ] Permanent smoke scenario added with this task id in owners
- [ ] cargo install --path . was run before claiming done

Depends on

• done .assign-wg-html-chat

Required by

• (none)

Log

• 2026-04-29T16:41:34.412127321+00:00 Task paused
• 2026-04-29T16:41:34.458041754+00:00 Task published
• 2026-04-29T16:41:56.803041576+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Careful Programmer has strong Rust track record (0.77 score, 312 tasks), and Careful tradeoff matches task's verification requirements (visibility logic, secret sanitization, live smoke testing, thorough validation).
• 2026-04-29T16:41:58.483994252+00:00 Spawned by coordinator --executor claude --model opus
• 2026-04-29T16:42:12.170020143+00:00 Starting investigation: locate wg html v2 viewer code, chat storage format, visibility model
• 2026-04-29T16:45:03.275091368+00:00 Design: --chat flag opts in to transcripts; --all (when paired with --chat) extends to internal transcripts; --public-only also filters chat tasks. Reuses src/chat.rs read_inbox_ref / read_outbox_ref with chat alias derived from task id. Sanitizer = small static regex set (sk-[a-zA-Z0-9]{20,}, OPENAI_API_KEY=..., paths under ~/.wg/secrets, generic api[_-]?key|secret|token = ...). Writing failing-first integration tests now.
• 2026-04-29T16:57:56.077503606+00:00 Task unclaimed: agent 'agent-1126' (PID 3634466) process exited
• 2026-04-29T16:58:04.640830615+00:00 Spawned by coordinator --executor claude --model opus
• 2026-04-29T17:07:37.649895401+00:00 Validated: cargo build clean (bin warns pre-existing); 26/26 integration_html tests pass (incl. 8 new chat tests); 16/16 html lib tests pass; smoke wg_html_chat passes 5/5 (default omits transcripts; --chat shows public; --chat --all shows all; --chat --public-only filters internal; sanitizer redacts sk-/OPENAI_API_KEY=/.wg/secrets paths)
• 2026-04-29T17:07:41.599473523+00:00 Live smoke against this project's graph: wg html --chat reports 'Showing 0 chat transcripts (28 hidden by visibility — pass --all to include them)' (correct — all chat tasks here are visibility=internal); --chat --all renders all chat task pages with Conversation sections (most empty because their UUIDs don't have inbox.jsonl). cargo install --path . succeeded.
• 2026-04-29T17:08:34.056661100+00:00 Committed: d699e3c80 — pushed to remote (wg/agent-1126/wg-html-chat)
• 2026-04-29T17:08:46.039033078+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
• 2026-04-29T17:11:02.469376065+00:00 PendingEval → Done (evaluator passed; downstream unblocks)