fix(planreview): anchor comments reliably and highlight what they point at

[tusharbhardwaj-bk]

Why

Anchored plan feedback reached the agent as raw <review_comment> XML instead of a card:

Three links, each reproduced end to end against the real functions before changing anything:

1. locateQuotedLineRange compared the reviewer's quote against raw markdown with String.includes. The quote comes from window.getSelection().toString() over rendered Plate output, so ** and backticks are already gone — '1. **Outbound email context**'.includes('1. Outbound email context') is false. Comments on formatted lines are the common case, not an edge case.
2. With no range located, formatPlanReviewComment omitted startIndex/endIndex.
3. parseReviewCommentContext required both indices, returned null, and the transcript fell back to printing the block verbatim — so the XML rendered as prose and the ```markdown fence became a monospace code block. apps/mobile had the identical gate.

Separately, CommentLeaf was registered but nothing ever applied a comment mark — a comment was real on the server and invisible in the document.

What changed

The prompt path

• planReviewAnchorText projects both sides onto the text a reader would select (strips block syntax, emphasis, code spans, links/images) before comparing. Only the comparison changes; quotedText still travels verbatim.
• The range is now optional in both parsers, so a future locate miss degrades to a card without a line range rather than to raw XML.
• A plan comment renders as its own card — plan title and a blockquote, not a workspace-relative path and a code block. The byline moved from the body into an author attribute.

The review surface

• Two-tier highlight: bright while composing, persistent amber once saved, hairline once resolved. Restored on reopen by re-locating the stored quote, so teammates' comments are highlighted too.
• Inline comment popover beside the selection (quote in the header, ⌘/Ctrl+Enter, Esc), replacing the box at the foot of the panel.
• One-click 👍 / remove quick labels.
• Two-way rail ↔ document linking.
• Width-gated Contents outline with per-section comment counts, via container query — no ResizeObserver.
• Send feedback only claims the primary slot once there is feedback to send.

Verification

123 tests passed across 8 scoped files; tsgo --noEmit clean for apps/web and packages/shared; apps/mobile is byte-identical to its pre-existing 76-error navigation baseline; lint and check-fork-markers.ts pass.

Two traps are pinned by tests against a real Plate editor, because either would make commenting look like a reviewer edit and send the agent a spurious edited-plan document:

• serialized markdown is byte-identical after applying a comment mark and after a draft mark;
• cancelling a draft restores the tree exactly.

Reviewer check on bkt3-dev: open a plan, select a bold list item, comment — the text should highlight and the transcript card should show the plan title and a quotation, with no <review_comment> text and no monospace block.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	11.4 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	5.7 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	5.7 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	50.4 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	11	—	21	✅
Claude	Total thread wire	—	11.5 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	5.7 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	5.7 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	51.2 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	11	—	21	✅

Baseline: unavailable · PR result: 57ef413 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 95.6 KiB
• Claude decoded thread snapshot: 96.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.