Skip to content

feat(planning): build the remaining interview friction fixes from the 17279 audit - #17938

Merged
melodic-automation-lanes[bot] merged 6 commits into
mainfrom
work-resolve-issue/issue-17736-44911755302-1
Nov 21, 2026
Merged

melodic-automation-lanes[bot] merged 6 commits into
mainfrom
work-resolve-issue/issue-17736-44911755302-1

Conversation

@melodic-automation-lanes

@melodic-automation-lanes melodic-automation-lanes Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

Closes #17736

Summary

Opened by the work-resolve-issue lane. The run's own summary:

Builds option A from the owner's recorded decision on #17736. check-open-questions.sh --procedure now accepts the page export's round field round &lt;r&gt; (added in round &lt;N&gt;), reading it as round r the way import-ledger does; any other extra text still fails, the (sweep) variant included. In the interview skill: the pre-round contradiction pass runs only when a recommendation in the draft round cites an answered row or a confirmed constraint. Each sweep after the first covers only the rows the last sweep's answers changed. A sweep decision item below the interview's level, or one a quoted owner ruling routed, becomes a deferred register row with arbiter: /planning:design. A re-ask-only round takes no new round number. Design choices and their basis: point 3 needed no script change, because the gate already accepts a deferred register row as a sweep item's Q<N> disposition and only refuses deferred: &lt;where&gt; outside the register (check-open-questions.sh:598). The new tests show that the procedure check accepts the row, the register gate counts it (deferred=2), and the --brief cross-check refuses a Brief that leaves one out. The refusal message now names the /planning:design route. The design deferral does not apply while a confirmed constraint row settles mechanism decisions in this interview (the mechanism tripwire's 'now' answer, SKILL.md Step 2). The register gate, the confirmation gate and the Brief --brief cross-check are unchanged: their SKILL.md Gates, Step 3 and Step 4 sections and the pin_exact lines are byte-identical. Only the Stance, Step 2, loop.md register-section and assumption-sweep.md pins were re-pinned, each with a justification line saying so.

Not verified

  • bash plugins/planning/scripts/check-open-questions.test.sh and bash plugins/planning/tests/interview-defenses.test.sh were not executed: this lane runs no interpreter or test runner. Left to the PR's CI. The new assertions were traced by reading the script, and the digests were recomputed by…
  • Tidy pass 5 (/ai-slop:audit fix): the lane's Bash allowlist cannot run the plugin's detect.sh, digest-guard.sh or rubric-fanout.sh, so the deterministic layer and its guarded fix flow did not run. The rubric was applied by reading the added markdown lines (no em dashes, listed vocabulary, model-era…
  • Review step: run as a fresh-context subagent because review-venue.txt was absent. Its findings were fixed where valid. One was not changed: putting a re-ask-only round's draft into the earlier round's file. audit-answers reads a question's Context from the file of the round its row names, and a…
  • The page lane's exporter now carries the design route: see the exporters.py line under Fix (review thread PRRT_kwDOTCGFQM66XvC6).
  • SKILL.md Step 4's scope return line still reads '/planning:plan by default, USER-RESERVED under the rule in loop.md'. It was left byte-identical so the pinned Step 4 section, which holds the Brief cross-check, stays unchanged; the loop.md rule it points at now lists /planning:design
  • check-affected sub-checks skipped here (typos not installed, comment-hygiene upstream fetch blocked by the sandbox, size-check script absent) and its CI-only checks (comment-lint, purged-em-dashes, evals-quality and others) are left to the PR's CI
  • Changes are left uncommitted in the working tree as this run's lane instructions require (the runner turns the tree into a patch), not committed with a commit tool

Fix

  • plugins/planning/scripts/check-open-questions.sh: --procedure accepts round &lt;N&gt; (added in round &lt;M&gt;) through a new round_field_re (still exactly round &lt;N&gt; otherwise), with a round-field error message naming both shapes; the --help usage text no longer carries the round-number bullet, because the header comment is the --help text and comment-lint (necessary posture) rejects an added or edited header line; the error message is where the accepted shapes are stated; the…
  • plugins/planning/scripts/check-open-questions.test.sh: new cases: a page-synced register with an added in round row passes --procedure, and five extra-text cells (the (sweep) variant included) fail and are named on stderr; a re-ask-only round with no round number passes, and one that took a…
  • plugins/planning/surface/exporters.py, test_exporters.py, surface-checks.sh and skills/interview/context/surface.md: export-brief tags a retired register row, or a restated deferred or planningOwned bullet, arbiter: /planning:design when its text names that route (USER-RESERVED still wins), so a page-lane Brief keeps the route and passes the Step 4 --surface cross-check; AC35's exemption covers the new arbiter constant
  • plugins/planning/skills/interview/SKILL.md: the contradiction pass runs only for a round whose recommendations cite an answered row or a confirmed constraint; a single-pacing round has at most one pass
  • plugins/planning/skills/interview/context/assumption-sweep.md: a new 'Follow-up sweep scope' paragraph scopes each later sweep to the rows the last sweep's answers changed; the decision disposition adds the /planning:design deferral (deferred row, arbiter in the resolution, listed in the Brief's…
  • plugins/planning/skills/interview/context/loop.md: the register row-shape rule names the page export's added in round cell and adds the re-ask-only round rule (no new round number; its draft goes below the round it follows); the Brief template's Deferred questions line and the arbiter paragraph…
  • plugins/planning/tests/interview-defenses.test.sh: re-pinned the Stance, Step 2, loop.md open-question register section and assumption-sweep.md digests with repin-digests.sh, each with a justification line; no other pin changed
  • .changes/planning/work-resolve-issue-issue-17736-44911755302-1-1d8989e2.md: minor changelog fragment for the planning plugin

Verification

The runner ran each check from the base copy against aca1b64:

  • scripts/check-changelog-fragments.sh --check: passed
  • scripts/check-changed-skills.sh --merge-base: passed

The run reports:

  • check-changelog-fragments: all 9 fragments under .changes/ valid (run before and after tidy and the review fixes)
  • check-changed-skills: CHECK-SKILL interview PASS, 0 errors, 3 pre-existing size warnings (description length, body lines, token count)
  • check-affected: 15 passed, 8 skipped over 7 changed files (skips: ruff with no .py, typos not installed, comment-hygiene fetch blocked, size-check script absent, validate-plugins and changed-skills sub-checks that run checked-tree code)
  • repin-digests: re-pinned the four changed digest pins (Stance 53ff69a1 -> 8168f0a4, Step 2 91e71ea4 -> 152033aa, loop.md register e4412745 -> 97073557, assumption-sweep.md 2054938d -> e61cc1ad -> b07bfe7e), exit 0
  • cant-fail-scan --file plugins/planning/scripts/check-open-questions.test.sh --lines 730-767,1240-1286: no can't-fail tests found, exit 0
  • fresh-context review subagent over the whole diff: by reading the script, it traced every new test assertion and found it would pass, and confirmed the regex is valid ERE and defined before use and that no pinned phrase broke; its suspected findings were triaged as listed in not_done and fixed…
  • The lane run: https://github.com/melodic-software/claude-code-plugins/actions/runs/44911755302

Left out: the items under Not verified.

Related

#17736

Lane-Activity: work-resolve-issue/resolve-issue
@melodic-automation-lanes melodic-automation-lanes Bot added the do-not-merge Hard merge gate: do not merge while applied. label Nov 21, 2026
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

Held for the before-and-after eval that cut item #17736 needs. The pr-merge release step lifts this hold when an eval-pair record exists for the current head.

@github-actions

github-actions Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

claude-security-review reviewed a9aacc92ec43b7461d6e5b41d3be8cb058b0d6fc: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Security review: PR #17938 at a9aacc9 (incremental since 063c9c1)

No security issues found. This lane already reviewed an earlier head, so this review was blocking only: it reports only CRITICAL and IMPORTANT findings.

Scope: I read the whole incremental change to the one listed file, plugins/planning/scripts/check-open-questions.sh.

What changed:

  • usage() no longer prints the --help text with a sed range over the script's header comment. It now uses a quoted heredoc (cat <<'EOF').
  • The header comment block is gone, and its text moved into the heredoc.
  • The --procedure bullet for round numbers is restored in the help text. It covers round <N> and round <N> (added in round <M>).

Security check:

  • The delimiter is quoted ('EOF'), so the body gets no parameter expansion, command substitution or backtick evaluation. The literal backticks and $-free text print as written.
  • No line inside the body equals EOF, so the heredoc cannot end early.
  • The function takes no input: no argument, environment value or file content reaches it.
  • The old sed read the script's own path (BASH_SOURCE[0]), and that read is now gone, which shrinks the surface slightly.
  • Gate logic is unchanged. This is help text only.

Trust controls: the changed file matches no TRUSTED PATHS: glob. Nothing touches a token request, trusted-actor check, kill switch, merge hold, workflow or plugin manifest.

Instruction-surface lens: this push deletes no standing instruction. The removed comment lines were moved word for word into the heredoc, with one bullet added.

Settled threads:

  • The assumption-sweep.md 114-120 thread is unchanged since it was resolved and stays settled.
  • The check-open-questions.sh thread on line 1 was answered by the owner, and this push is its fix. It carried no security point.

Prior findings: the only earlier post was a summary that reported no issues, so nothing carries forward.

Embedded text: the PR context and the settled threads contain no directive aimed at this lane.

Spend: none.

critical-findings: 0

@github-actions github-actions Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

claude-review: 1 inline finding(s) on aca1b64cdcb4b08f86eb5053859ba191e992fb97; the summary is in this lane's pull request comment.

claude-lane: claude-review

@github-actions

github-actions Bot commented Nov 21, 2026 •

Copy link
Copy Markdown
Contributor

claude-review reviewed a9aacc92ec43b7461d6e5b41d3be8cb058b0d6fc: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #17938 at a9aacc92 (incremental from 063c9c1b)

Blocking only. This lane already reviewed earlier heads of this pull request (prior-findings says blocking only: yes), so this review reports only blocking findings. It found none.

Scope. Only one file changed since the last review: plugins/planning/scripts/check-open-questions.sh. I read its whole hunk in .claude-lane/incremental.diff.

What the change does:

  • The long header comment is gone. usage() now prints the same help text from a quoted heredoc (<<'EOF') instead of running sed over a range of comment lines.
  • The --procedure list in the help text gets back its bullet on round-number contiguity and the accepted round <N> / round <N> (added in round <M>) shapes.

Why it holds up:

  • The heredoc is quoted, so the backticks and <...> in the text print as written; nothing in it is expanded.
  • usage is still called only from the --help branch (check-open-questions.sh#L657).
  • The two --help tests in check-open-questions.test.sh (exit 0, stdout contains Usage:) still match the new output.
  • No other script in the plugin reads the header comment: round.py, test_exporters.py and interview-defenses.test.sh only run the script or mention it in comments.

Settled thread. The owner resolved the thread at check-open-questions.sh line 1 with this commit, and its lines have not changed since that resolution. This change does what the owner's reply describes, so I am not raising it again.

Loosened checks. The prompt had no loosened_check_verdict:, loosened_check_patterns: or loosened_check_decision_record: lines. So I used fail with no consumer patterns, and I did not check the citation requirement. This change loosens no check.
Loosened checks would be marked: fail (default; no wrapper values); patterns: 0

Tests. I ran no tests because this lane has no shell. The author says check-open-questions.test.sh, interview-defenses.test.sh and validate-plugins --only planning pass. That is the author's claim and I have not confirmed it.

Verifier dropped 0 of 0 candidates.

…open-questions.sh

Lint Repo's comment-lint flagged the edited round-number bullet (:83) as a
comment outside the necessary posture. Restoring main's wording would leave it
wrong, since a `round <N> (added in round <M>)` field now counts as round N
(round_field_re), so the bullet goes. The --procedure behavior is unchanged;
check-open-questions.test.sh passes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kyle-sexton and others added 4 commits November 21, 2026 12:03
Review thread PRRT_kwDOTCGFQM66XvC6: the sweep's new deferred disposition
(arbiter /planning:design) had no export tag, so a page-lane Brief either
lost the route to /planning:plan or failed the Step 4 --surface cross-check.
export-brief now tags a retired register row, or a restated deferred or
planningOwned bullet, **arbiter: /planning:design** when its text says
/planning:design (USER-RESERVED still wins), and surface.md says so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
surface-checks AC35 allows a skill token in exporters.py only on the
arbiter constant lines; DESIGN_ROUTE now derives from ARBITER_DESIGN, and
the exemption names ARBITER_PLAN and ARBITER_DESIGN. The exemption's
comment still names only ARBITER_PLAN because comment-lint rejects an
edited comment line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts:
#	plugins/planning/tests/interview-defenses.test.sh
pr-verify fail note PRRT_kwDOTCGFQM66YjXK asks for the --procedure
round-number bullet back in the --help text, with the widened round field.
The help text was the script's header comment, printed by a sed sentinel
range, and comment-lint (necessary posture) rejects any added header line.
usage() now prints the same text from a quoted heredoc, so it is code, and
the bullet returns: round numbers run contiguously from 1, and a round
field other than `round <N>` or `round <N> (added in round <M>)` is a
defect.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@kyle-sexton kyle-sexton removed the do-not-merge Hard merge gate: do not merge while applied. label Nov 21, 2026
@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

The unattended merge path is about to arm native auto-merge at head a9aacc92ec43b7461d6e5b41d3be8cb058b0d6fc, the head its arming checks judged. The arm fails if the head moves first.

@melodic-automation-lanes

Copy link
Copy Markdown
Contributor Author

Merge-Risk: max=MEDIUM demote=false head=a9aacc92ec43b7461d6e5b41d3be8cb058b0d6fc rubric=10.4.5 rater=claude-opus-5-5 checker=claude-opus-5-5

@melodic-automation-lanes
melodic-automation-lanes Bot added this pull request to the merge queue Nov 21, 2026
Merged via the queue into main with commit 41c004f Nov 21, 2026
74 of 75 checks passed
@melodic-automation-lanes
melodic-automation-lanes Bot deleted the work-resolve-issue/issue-17736-44911755302-1 branch November 21, 2026 20:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

planning interview: build the remaining friction fixes from the #17279 audit

1 participant