docs(topics): commit two uncommitted program Briefs for durability

[kyle-sexton]

No linked issue — durability commit for two existing contracts; program disposition is ruled separately.

docs/topics/fresh-eyes-checkpoint-audit/PLAN.md (Brief locked 2026-07-19 for #304, Plan never filled) and docs/topics/underspecification/PLAN.md existed only as untracked working-tree files — a locked contract one git clean -fdx from loss, and repeatedly dismissed in handoffs as "untracked dirt NOT ours." Committed verbatim, no content changes.

Related

• program: fresh-eyes checkpoint audit — tag skill actions for same-context bias and retrofit delegation #304 (fresh-eyes-checkpoint-audit program — park/resume ruling pending)
• Drift-audit curation pass 2026-07-20 (this session)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8aec984627

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 2m 13s —— View job

---

PR Review

• Gather context and read changed files
• Review docs/topics/fresh-eyes-checkpoint-audit/PLAN.md
• Review docs/topics/fresh-eyes-checkpoint-audit/design/design-resolution.md
• Review docs/topics/underspecification/PLAN.md
• Evaluate Codex reviewer comments
• Post final review summary

---

Purpose: Durability commit for two topic Briefs that existed only as untracked working-tree files. No content was changed — the files are committed verbatim. The PR description is accurate; disposition of the programs is deferred.

The durability rationale is sound. The files contain valuable locked contracts and design decisions that would be lost to git clean -fdx. There is no objection to committing them as-is. What follows are two issues that need correction before either Brief is acted on, confirmed independently from the Codex review.

---

Finding 1 — docs/topics/underspecification/PLAN.md Brief is stale (confirmed: P2)

The Goal section asserts:

The concept … is never named in the repo, so the term neither triggers the right skill nor appears in skill framing …

This is no longer true. I verified against origin/main:

• plugins/planning/skills/interview/SKILL.md — "underspecified" in both the description trigger and the Purpose paragraph
• plugins/planning/skills/design/SKILL.md — "underspecification" in the Purpose paragraph; "underspecified types" in the step list
• plugins/planning/skills/prd/SKILL.md — "underspecified" in the routing table
• plugins/planning/CHANGELOG.md — documents the vocabulary addition as a completed release

The acceptance criteria is met. Any session that resumes from this Brief will either redo already-merged edits (if the agent doesn't check origin/main first) or spend time determining that the work is complete and the Brief is stale.

Recommendation: Before this topic slice is closed out, update the Brief's Goal section to reflect that the vocabulary edits were merged, and either mark the Plan section done or prune the slice with a pointer to the merged PR. This is not a blocker for the durability commit itself but should be addressed before the next session picks up this topic.

Fix this →

---

Finding 2 — docs/topics/fresh-eyes-checkpoint-audit/PLAN.md line 349: re-anchor skill count contradicts Phase 2 table (confirmed: P2)

The [EXEC-SHAPE] bullet at line 349 says:

re-anchor's four skills = one PR; one PR per plugin otherwise

But the Phase 2 table at line 200 says:

shared plugins/re-anchor/context/re-anchor-audit-correct.md + ALL 11 skills referencing it

I verified: plugins/re-anchor/skills/ contains exactly 11 skills, and all 11 of them reference re-anchor-audit-correct.md. A worker that follows the [EXEC-SHAPE] bullet and scopes the PR to four skills would leave 7 skills without a per-skill declaration anchored in their own SKILL.md. The design contract explicitly requires per-skill anchoring because "the checker is generic and cannot assume a plugin layout" — so without those per-skill anchors the check-18 gate is vacuous for the unretrofitted skills.

The "four re-anchor correctors" phrasing also appears in the Deferred questions section (line 93) in the context of whether -deep siblings are warranted — this appears to be a separate concept (a subset of correctors with escalation-criteria questions) that was inadvertently conflated with the full Phase 2 PR scope in the [EXEC-SHAPE] bullet.

Recommendation: Correct line 349 to read ALL 11 skills = one PR before Phase 2 execution begins. The authoritative spec is the Phase 2 table; the [EXEC-SHAPE] bullet should match it.

Fix this →

---

docs/topics/fresh-eyes-checkpoint-audit/design/design-resolution.md — no issues

The design resolution is internally consistent. The declared-pattern contract (delegation pattern + exemption directive grammar + check-18 semantics table + scan mechanics) is clearly specified and normative. The research base citations follow the repo's verified YYYY-MM-DD style. The spec-ownership decision (mechanical spec ships inside skill-quality) is well-motivated and aligns with the published-plugin constraint. No issues found.

---

Overall

The durability commit achieves its stated goal. Both findings above come from content that pre-dates the commit and should be corrected before the programs resume rather than in this PR. The Codex P2 assessments for both findings are correct.

[claude]

Claude finished @kyle-sexton's task in 1m 10s —— View job

---

PR Review

• Gather context and identify latest commit
• Review fix in docs/topics/underspecification/PLAN.md
• Review fix in docs/topics/fresh-eyes-checkpoint-audit/PLAN.md line 349
• Post final review summary

---

The latest commit (56351c5) addresses both P2 findings from the previous review round. Changes are correct.

Finding 1 — underspecification Brief staleness (resolved)

A COMPLETED banner was prepended at the top of the file:

> **COMPLETED (recorded 2026-07-21).** Every acceptance criterion below shipped before this
> Brief was committed: the planning skills (interview, design, prd) and the planning
> CHANGELOG name "underspecification"/"underspecified" as first-class vocabulary. Preserved
> as the program's historical contract; no further work.

The stale Goal section body is intentionally left in place under the banner as historical record, which is appropriate for a durability commit. A future session picking up this topic will see COMPLETED immediately and stop. No issues.

Finding 2 — re-anchor skill count contradiction (resolved)

The [EXEC-SHAPE] bullet at line 349 was corrected from:

re-anchor's four skills = one PR

to:

re-anchor's shared spoke + all 11 skills = one PR (matching the Phase 2 row
above; an earlier draft said four — corrected at rescue time)

The inline parenthetical explaining the prior "four" is useful for anyone reading git history and tracking down the discrepancy. The corrected bullet now matches the Phase 2 table and the actual skill count in plugins/re-anchor/skills/. No issues.

---

Overall

Both prior findings are resolved. The PR achieves its stated purpose — durability commit for two locked contracts — and the follow-up corrections land in this same branch before merge. No remaining issues.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 56351c51e2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[claude]

Claude finished @kyle-sexton's task in 1m 9s —— View job

---

PR Review

• Read current branch files
• Review latest Codex finding (fleet recheck, commit 2c57486)
• Verify prior findings remain resolved
• Post final review summary

---

The latest commit (2c57486) directly addresses the third Codex P2 finding from the review of 56351c5.

Finding 3 — fleet recheck claim (resolved)

The "Open questions" note at line 325 previously ended with:

Until then the recheck mechanism is the repo-fleet-hygiene audit.

Commit 2c57486 replaces that with:

(Rescue-time correction: no interim recheck mechanism exists — the repo-fleet-hygiene audit inventories git/GitHub state only and never invokes the skill gate, so until the trigger fires the WARN-clean invariant is unmonitored; the resuming implementer picks the recheck mechanism.)

The Codex reviewer asked for either CI wiring or an honest description of what the actual checker is. The correction takes the second path: it explicitly states no interim mechanism exists, correctly characterises repo-fleet-hygiene:audit as a git/GitHub inventory tool (not a skill-gate runner), and defers mechanism selection to the implementer at trigger time. The DEFERRED-with-trigger framing is already established in this Brief; the correction fits the pattern cleanly. No issues.

---

All findings from this review thread resolved:

Finding	Introduced in	Resolved in	Status
Underspecification Brief stale Goal section	8aec984	56351c5 (COMPLETED banner)	✓
Re-anchor skill count four vs ALL 11	8aec984	56351c5 (corrected [EXEC-SHAPE] bullet)	✓
Fleet recheck claim pointed at wrong tool	8aec984	2c57486 (rescue-time correction)	✓

No remaining issues. The PR achieves its stated purpose — durability commit for two locked contracts — with all three review-round corrections landing on the same branch before merge.