[kyle-sexton]

No related issue: step 3 of the owner-approved auto-ready plan (.work/auto-ready/PLAN.md, local).

Summary

/work-items:work now readies its pull request right after create --pushed, so its PRs leave draft without a manual flip once /source-control:pull-request ready's checks pass.

Fix

• work/SKILL.md, "Open the PR": right after create --pushed, the orchestrator dispatches /source-control:pull-request ready into the persisted worktree at the strong-tier model, because ready runs the code and security reviews. ready resolves the PR from the checked-out branch, so the brief names the PR create returned and stops on a mismatch. The dispatch is skipped when the consuming project reserves the ready flip for a person. A refusal (a hold, a superseded item, a failed gate) leaves the draft. The reason goes into the run's report, and nothing lifts the hold or flips the PR another way.
• Eval cases 8 (ready dispatched after create --pushed) and 9 (a held PR stays a draft when ready refuses).
• Changelog fragment (minor).

Verification

• scripts/check-affected.sh --tests: validate-plugins failed once on a timing assertion in a mod test that this diff does not touch ("runs each of the worst accepted extra patterns in time linear in the text length", 508 ms against a 225 ms bound). scripts/validate-plugins.sh --only work-items then passed on rerun (exit 0). Every other check passed.
• skill-quality check-skill.sh plugins/work-items/skills: work PASS (0 errors). check-evals-quality.sh on work's evals: PASS, 0 warnings.
• After merging main (one conflict in work/SKILL.md, composed: main's auto-merge rewording plus this PR's step): scripts/check-affected.sh --tests 7 passed, 2 skipped; cross-plugin drift and shared-copy checks clean; check-evals-quality.sh PASS.
• Not run: the model-graded eval cases, and the plan's live run on a test issue.

Related

• Builds on feat(source-control): refuse a held or superseded draft in pull-request ready #7345 (ready refuses held and superseded drafts) and docs: let the session that opened a PR flip it through pull-request ready #7346 (the AGENTS.md carve-out that lets the opening session flip its own PR); both are merged.
• feat(work-items): ready each PR after create in ship run #7350 adds the same step to /work-items:ship run (skills/ship/actions/run.md).

🤖 Generated with Claude Code

https://claude.ai/code/session_01W6SjPvY7DmXBGzboeCq8nj

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-13T18:27:01.562176Z	036fbc6	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[claude]

Claude finished @kyle-sexton's task in 1m 59s —— View job

---

Security review: no issues found

• Gather context (PR diff, changed files)
• Check for instruction-surface deletions
• Review security-relevant surface
• Adversarial validation of any candidate findings
• Post final summary

Scope reviewed: the full diff (gh pr diff 7347) across all three changed files:
.changes/work-items/feat-work-ready-after-create-3aeab667.md,
plugins/work-items/skills/work/SKILL.md,
plugins/work-items/skills/work/evals/evals.json.

What changed: a prose-only edit to one bullet in work/SKILL.md ("Open the PR"), adding a
"Then ready it" instruction that has the orchestrator dispatch /source-control:pull-request ready right after create --pushed, plus two new eval cases and a changelog fragment. No code,
no shell, no new tool surface — just skill-body text that an agent reads as instructions.

Instruction-surface check: the edited bullet is additive, not a deletion/narrowing of a
standing guardrail. The second commit tightens the new instruction further: it adds an explicit
stop when ready resolves a PR number different from the one create --pushed returned, and
skips the dispatch outright when the consuming project reserves the ready flip for a person. The
automated-ready behavior itself is the documented, already-merged exception in this repo's
AGENTS.md (the session that opened a PR may flip it via /source-control:pull-request ready
run to completion); this change operationalizes that existing, reviewed policy inside the work
orchestrator rather than introducing a new agent-authority grant. No Gate 0 class (irreversible
action, secret handling, data integrity, external publication, legal compliance, agent authority)
is weakened by this diff — holds, refusals, and the "never flip or merge another way" constraints
are preserved and reinforced.

No logic flaws, injection surfaces, credential handling, or authorization gaps found in this
diff.

claude-lane-summary: lane=claude-security-review sha=036fbc63f65e96824d5dd1f1e235a42dbfbb090b run=38446445243

[claude]

Claude finished @kyle-sexton's task in 3m 10s —— View job

---

Code review — /review:code-review

• Gather context (PR diff, changed files, CLAUDE.md/AGENTS.md, REVIEW.md)
• Check skip-gate conditions
• Review work/SKILL.md changes
• Review evals/evals.json changes
• Review changelog fragment
• Post findings

Scope. Reviewed the full diff in .claude-lane/pr.diff: the new "Then ready it:" clause in plugins/work-items/skills/work/SKILL.md (Step 5, "Open the PR"), the two added eval cases (8, 9) in evals/evals.json, and the changelog fragment.

What I checked and ruled out:

• ready's own contract (reference/ready-for-review.md) resolves the PR from the current branch, not from an argument — the new text correctly has the brief name the expected PR number for the dispatched agent to self-check against, rather than claiming ready takes a PR-number argument it doesn't support.
• Dispatching "into the persisted worktree" (no --worktree flag passed to ready) matches the existing pattern already used for monitor and the post-green review pass dispatch a few sentences later in the same paragraph — consistent, not a new mechanism.
• ready never merges the PR (only flips draft→ready), so the "Never-merge boundary" step later in this file is unaffected.
• The two new eval cases follow the existing schema (id/name/prompt/expected_output/files/expectations), and narration: true is the correct, already-used opt-out for fictional worktree paths (/work/wt-731, /work/wt-813) that resolve to no shipped fixture (plugins/skill-quality/scripts/check-evals-quality.sh Q4).
• The changelog fragment is correctly minor-bump, work-items is listed in scripts/fragment-plugins.txt, and its ### Changed shape matches other fragments.
• No docs/standards/README.md row applies (paved-path surface is absent), so nothing to flag there.
• Author-claimed verification (test runs, check-skill.sh, check-evals-quality.sh results) in the PR body is unverified by me — I did not have tooling access to re-run them in this lane.

No blocking or high-signal findings. This is a prose-only skill change plus eval additions; it's internally consistent with the surrounding file's established patterns and with the already-merged policy basis (#7345, #7346).

claude-lane-summary: lane=claude-review sha=036fbc63f65e96824d5dd1f1e235a42dbfbb090b run=38446445700

[github-actions]

claude-security-review has reviewed this pull request through 036fbc6; a later push is reviewed from there.

[github-actions]

claude-review has reviewed this pull request through 036fbc6; a later push is reviewed from there.