fix(web): show the whole last turn in the Claude response viewer's brief view

[shenlvkang-collab]

Summary

Follow-up to #369. The eye button's brief view rendered data.text, which is one row: the last assistant row of the transcript. A Claude answer is a median of 3 model messages (p90 11) split around tool calls, so the brief view usually showed the tail of an answer ("Done.", "Let me look.") while the substance appeared only after More. The full view was fine, which is why the brief one read as broken by comparison.

• The brief view now asks ?context=turn. The Claude reader answers with the assistant messages of the last answered turn (selectLastAnsweredTurn() in response-viewer-transcript.ts: the highest turn that has an assistant row, so a prompt queued after the answer does not blank the view), and the frontend renders them exactly as the full view renders that turn — one badge, then continuation segments, gated on the numeric turn as before.
• data.text is unchanged in every context — still the last assistant row, never messages.at(-1) — because agent pollers hash it (skills/codeman/preamble.sh).
• Readers that emit no turns (Codex, the pane parser, DeepSeek, an older server) return text only for context=turn, and the brief view keeps its single card for them. The terminal-buffer fallback is untouched.

Docs: skills/codeman/reference/endpoints.md (new row), docs/architecture-invariants.md (the viewer paragraph). Changeset: patch.

Test plan

• test/routes/session-routes-claude-last-response.test.ts: context=turn returns the last answered turn's assistant rows (not the queued prompt after it) with text still the last row; empty list before the first answer.
• test/response-viewer-last-turn.test.ts (new): the pure selector (highest answered turn, no numeric turns → empty, turn 0 kept) and the brief view in jsdom (requests context=turn, one badge + segments, falls back to text, ignores user/blank rows).
• test/response-viewer-turn-segments.test.ts, test/response-viewer-transcript.test.ts unchanged and green.
• tsc --noEmit, eslint, prettier.

🤖 Generated with Claude Code

https://claude.ai/code/session_01McLWqCWBuQYGuPMScb4Aou

[Ark0N]

Merged, thank you! This is a genuinely good catch: data.text is one row and a Claude turn is a median of 3, so the eye button was showing the "Done." tail of an answer whose substance was in the rows above it. The measurement behind it (285 real transcripts, p50 2 rows / p90 13) is exactly the kind of evidence that makes a change like this easy to accept, and I appreciated that you pinned data.text for agent pollers with a test rather than just asserting the contract held.

One thing applied on master at merge time (02b0e278): the view opened at body.scrollTop = 0, which was right when it was a single card holding the last row, but with the whole turn rendered the top of the scroller is the turn's first narration line. loadFullContext() already scrolls to the bottom of that same turn, so the two views disagreed about which end mattered. It is now body.scrollTop = turnMessages.length > 1 ? body.scrollHeight : 0, so a multi-row turn opens at its newest text and a single card still opens at the top, with a test covering both.

The remaining review notes (the duplicated numeric-turn continuation gate, a mid-turn queued prompt fusing two answers under one badge, and no route test pinning the non-Claude context=turn behaviour) are cleanups rather than blockers. Happy to take any of them as a follow-up if you want them, otherwise they go in a later pass.

Shipping in the next release.