fix(planning): interview page batch for eleven page issues

[kyle-sexton]

Requested by Kyle · project thread

Closes #5801
Closes #5815
Closes #5916
Closes #5914
Closes #5922
Closes #5917
Closes #5908
Closes #5910
Closes #5909
Closes #5817
Closes #5913
Refs: #5911 (not started: work stopped before it)

Summary

This is a batch of /planning:interview page fixes. There is one commit per issue, and each commit subject names its issue. All changes are in plugins/planning and share one changelog fragment (bump: minor).

• planning: refuse a recommendation change that leaves the alternatives unreviewed #5801
  • Before: a reply or revise could change the recommendation and leave the alternatives unchecked.
  • After: such an op must carry the revised alternatives or keepAlternatives: true (CLI --keep-alternatives). Otherwise it is refused. reply also accepts alternatives, and the history line records rewritten or kept.
• planning: show the whole recommendation in the interview page's Accept choice #5815
  • Before: the Rec choice showed only the first sentence of the recommendation.
  • After: it shows the whole recommendation. The Accept all list keeps its one-sentence preview.
• planning: fix the duplicated "Recommendation revised." banner and retire it once seen #5916
  • Before: the revised banner could say "Recommendation revised." twice.
  • After: it says it once. A change with no why now stores revised: true. Once the user has opened the question since the revision, the banner shrinks to a "Revised" chip.
• planning: let Claude move a question to another group and set the group order #5914
  • Before: a question could not move to another group, and groups could not be reordered.
  • After: revise --group moves a question in place and logs a history line. The new group-order op reorders groups.
• planning: name the stale ancestor, not the direct parents, when the interview page marks a question upstream-pending #5922
  • Before: the upstream-pending chip and banner named the direct prerequisites.
  • After: they name the stale ancestor(s), from a new derived upstreamStale field that is never persisted. The banner links them with a reconfirm hint, and the row reads as waiting, not locked.
• planning: lay out multiple "needs" chips in one row and mark met prerequisites in text #5917
  • Before: each needs chip took its own line, and met vs unmet differed only by colour.
  • After: the chips share one wrapped row, and a met prerequisite carries a check mark labelled "answered".
• planning: say "decided after you answered" on a stale question and show the review count in the meter #5908
  • Before: the stale banner always said "Qn changed".
  • After: it says "was decided after you answered this" when that is the case. The meter adds "· N to review", which opens the Review filter.
• planning: make Reconfirm on a stale question say exactly what it records #5910
  • Before: a kept accept offered Reconfirm even after its recommendation changed.
  • After: no Reconfirm is offered in that case. a arms Rec instead.
• planning: after a changed answer, take the user to the dependents it made stale #5909
  • Before: after a save that made other answers stale, the page moved on with no sign of them.
  • After: the page lands on the first newly stale question and switches Open to Review if needed. The toast names the stale questions, and n visits the rest first.
• fix(planning): interview page does not show Claude's reply to a question the user asked #5817
  • Before: the latest Claude reply was shown only in the latest-reply box.
  • After: it also stays in the Thread.
• planning: keep the question body visible when the note or conflict box grows #5913
  • Before: a long note or the conflict box could squeeze the question body to a strip.
  • After: the body keeps at least 45% of the pane, and the answer panel scrolls.

Fix

• planning: refuse a recommendation change that leaves the alternatives unreviewed #5801: round.py gains require_alt_review and checked_alts. op_reply and op_revise use them, and there is a new CLI flag. The ops and questions schemas are updated, along with surface.md and README.md. Existing rec-change call sites in tests and the harness now pass --keep-alternatives.
• planning: show the whole recommendation in the interview page's Accept choice #5815: choicesOf gives the Rec row the full recommendation text, and a CSS rule unclamps the Rec row. A follow-up test commit points the ui_a two-line clamp check at a non-Rec option.
• planning: fix the duplicated "Recommendation revised." banner and retire it once seen #5916: revised stores the why or true, and the schema allows a string or boolean. The page has recRev, a per-question revSeen in browser storage, and the .revchip.
• planning: let Claude move a question to another group and set the group order #5914: op_revise takes group and checks it exists. It warns on stale group summaries. The new op_group_order requires every group id exactly once. The schema, surface.md and README.md are updated.
• planning: name the stale ancestor, not the direct parents, when the interview page marks a question upstream-pending #5922:
  • server.py: derive_states now also returns stale ancestors per upstream-pending question (breadth-first, nearest first, ties by id). read_state exposes them as upstreamStale.
  • Page: a waitOn() helper feeds the chip and the banner, and a .qbtn.waiting style replaces the dim.
• planning: lay out multiple "needs" chips in one row and mark met prerequisites in text #5917: a .needsrow wrapper and a .metmark icon with role="img" and aria-label.
• planning: say "decided after you answered" on a stale question and show the review count in the meter #5908: firstDecided() drives the per-prerequisite banner wording. A #revBtn meter count uses the Review filter's predicate.
• planning: make Reconfirm on a stale question say exactly what it records #5910: reconfirmOf returns null for a kept accept when a rec-change history line (pageSeq) is at or after the accept's seq, or has a newer rev for a terminal record. The shortcut help and the surface.md Upstream changed paragraph are updated.
• planning: after a changed answer, take the user to the dependents it made stale #5909: send() compares stale and upstream-pending sets before and after its refresh. It sorts the newly stale questions by dependency depth and switches the filter when needed. nextNeeding serves the queued stale questions first.
• fix(planning): interview page does not show Claude's reply to a question the user asked #5817: the Thread no longer filters out the latest Claude reply. In historyOf, a reply in the same second as the event it answers now sorts after that event.
• planning: keep the question body visible when the note or conflict box grows #5913:
  • main.detail rows are minmax(45%,1fr) minmax(0,auto).
  • section.answer scrolls.
  • grow() caps the note against the pane.

Overlap with draft PR #7036 (fix/6946-interview-yes-accepts-rec): none. That PR only touches plugins/planning/skills/interview/SKILL.md and its own .changes fragment, and this PR changes neither file.

Verification

• Python: python3 -m unittest test_round test_schema test_server passed (401 tests, 1 skipped) after the planning: let Claude move a question to another group and set the group order #5914 commit. The new TestUpstreamStaleNamesTheCause (planning: name the stale ancestor, not the direct parents, when the interview page marks a question upstream-pending #5922) and TestRecChangeMarksUpstream were rerun after planning: name the stale ancestor, not the direct parents, when the interview page marks a question upstream-pending #5922. Later commits change only the page and its tests.
• Lint: scripts/run-ruff.sh format / check is clean on plugins/planning/surface. node --check passes on the page script after each change.
• Full suite: surface.test.sh was last run in full after planning: lay out multiple "needs" chips in one row and mark met prerequisites in text #5917 (before planning: say "decided after you answered" on a stale question and show the review count in the meter #5908): PASS=491, FAIL=2.
  • The AC34 "user name found" check fails here because the runner is root and server.py contains the word "root". It fails the same way on pristine main (479/1), so it is an environment artifact.
  • The other failure was a ui_a overflow flag on the visually hidden "(answered)" span. It was fixed in the planning: lay out multiple "needs" chips in one row and mark met prerequisites in text #5917 commit by switching to an aria-label icon.
  • The full suite was not rerun after planning: say "decided after you answered" on a stale question and show the review count in the meter #5908 to planning: keep the question body visible when the note or conflict box grows #5913.
• Browser phases: the new ui_page phases 1 to 8 were run on their own after each later change, and the last run gave PASS=37, FAIL=0. They cover planning: show the whole recommendation in the interview page's Accept choice #5815, planning: fix the duplicated "Recommendation revised." banner and retire it once seen #5916, planning: lay out multiple "needs" chips in one row and mark met prerequisites in text #5917, planning: say "decided after you answered" on a stale question and show the review count in the meter #5908, planning: after a changed answer, take the user to the dependents it made stale #5909, planning: make Reconfirm on a stale question say exactly what it records #5910, the two-level case of planning: name the stale ancestor, not the direct parents, when the interview page marks a question upstream-pending #5922, fix(planning): interview page does not show Claude's reply to a question the user asked #5817 and planning: keep the question body visible when the note or conflict box grows #5913.
• Fails before the fix: phase 8 (planning: keep the question body visible when the note or conflict box grows #5913) fails on the previous page (body share 0.25 with the long note, 0.06 with the conflict box). Phase 7 (fix(planning): interview page does not show Claude's reply to a question the user asked #5817) asserts the reply in #thread, which the old de-duplicating filter removed.
• Changelog: scripts/check-changelog-fragments.sh --check and --check-required origin/main both pass.
• Environment note: under heavy load, Chrome sometimes failed to launch from a long TMPDIR path. Runs used TMPDIR=/tmp/claude-0/ivs.

Related

• planning: keep "Claude is working on Qn" visible while a status is set #5911 (Claude line: status and working text side by side) was not started; work stopped before it.
• Every issue here carries needs-human (a work-loop C3 ratification gate), and planning: let Claude move a question to another group and set the group order #5914 also needs Kyle's attend-queue approval. The labels were left for Kyle.
• Draft PR fix(planning): word interview questions so yes accepts the recommendation #7036: no overlap (see Fix).

🤖 Generated with Claude Code

https://claude.ai/code/session_01L8KdG7fWg2YrErVYWtohmR

---

Generated by Claude Code

[kyle-sexton]

ci-status is red because this PR is a draft, not because of the change. On a draft every lint and test lane skips (lint-repo, test-bash, test-node, test-python, check-skills, check-plugins), and the aggregate counts a skip as not passing. See .github/workflows/pr-require-checks.yml:507-513.

The lanes run once the PR is marked ready. That flip waits for Kyle, because it also starts the review lanes. Until then, the local runs listed under Verification in the PR body are the signal.

---

Generated by Claude Code