fix: session-close drops DONE when the anchor is still in progress

[bborbe]

Problem

/vault-cli:session-close § Closer panel prescribes ⚪ DONE for every clean verdict — mode 1, mode 2, and the goal-anchored case. Its own mode gate deliberately permits reaching those verdicts when the anchor task was resolved this session but still reads in_progress on disk: "a carried-forward in_progress task does not block them."

~/.claude/CLAUDE.md names exactly that an anti-pattern — "⚪ DONE while the anchored task is still in_progress" — and requires 🔵 READY / 🟡 WAITING / 🔴 BLOCKED in its place.

The two rules cannot be reconciled by wording. § Closer panel says "verbatim, no rewording", so a session that reaches mode 1 or 2 with a carried-forward anchor has no legal closer to emit: following the command breaks the memory rule, and following the memory rule breaks the command's own verbatim requirement.

Fix

One paragraph after the § Closer panel intro. It keeps the block's body unchanged and swaps its ⚪ DONE line for the panel the anchor's own state calls for — 🟡 WAITING with later (on <trigger>): when every open item is clock-bound, 🔵 READY when the next step is the operator's. Only the anchor's own status licenses ⚪ DONE.

Closing a session is not the same claim as the task being finished.

Evidence

Measured 2026-10-14 and 2026-10-15: two consecutive runs reached a clean verdict with a carried-forward anchor and had to improvise the panel against this section's own verbatim rule, because no block below offered one.

Verification

make precommit — PASS (ready to commit). CHANGELOG structure OK, golangci-lint 0 issues, go vet clean, all scenario/rule suites green, 0 vulnerabilities and 0 secrets.

[ben-s-pull-request-reviewer]

No automated checks needed for a markdown-only change. Let me review the diff.

Reviewing commands/session-close.md:

The added paragraph (lines 711-712) states: when a carried-forward in_progress anchor keeps the clean verdict (modes 1 or 2), the ⚪ DONE line is replaced with the panel the anchor's own state calls for — 🟡 WAITING when items are clock-bound, 🔵 READY when the next step is the operator's.

The logic is sound: ~/.claude/CLAUDE.md forbids ⚪ DONE while the anchor is in_progress, but the mode gate lets a resolved anchor reach clean verdicts. The verbatim-rule prevents rewording the closer blocks, so the fix must substitute the right panel for the ⚪ DONE line. The two observed cases (2026-10-14 and 2026-10-15) are cited.

Reviewing CHANGELOG.md:

The Unreleased entry accurately describes the bug, the mechanism of the fix, what's in scope and out of scope, and the measured observation. Consistent with the code change.

Correctness: The fix is correct. The closer panel rules (⚪ DONE for clean verdict) conflicted with CLAUDE.md's anti-pattern rule (⚪ DONE while anchor is in_progress). The resolution is correct: keep the clean verdict (nothing is blocking), but substitute the panel that matches the anchor's actual state.

{
  "verdict": "approve",
  "summary": "Bugfix corrects the session-close closer panel: when a carried-forward in_progress anchor reaches the clean verdict, the ⚪ DONE line is replaced with the appropriate 🔵 READY or 🟡 WAITING panel per CLAUDE.md's own anti-pattern rule. CHANGELOG and code are consistent. No functional code changed.",
  "comments": [],
  "concerns_addressed": [
    {
      "concern": "correctness: session-close closer panel logic change — should show panel based on anchor state instead of always showing DONE",
      "disposition": "addressed",
      "detail": "commands/session-close.md:711-712 — clean verdict is preserved but the ⚪ DONE line is substituted with 🔵 READY or 🟡 WAITING when the anchor is still in_progress; CHANGELOG.md documents the fix in Unreleased"
    }
  ]
}