feat: add create-pr skill

[byapparov]

Closes #24

Summary

• add the portable create-pr skill for evidence-backed reviewer-ready PR drafts
• require a resolved branch route, promotion path, and linked issue before PR mutation
• provision documented issue/project metadata (for example Status = Development) only with an explicit mapping and confirmation
• keep authored claims, test sources, executed checks, configuration, and deployed state distinct
• add behavioral evals and forward-test evidence for rollout/test claims, unrelated local work, branch strategy, and missing issue links
• normalize public Markdown documentation branding to aictrl.dev

Verification

• python3 /home/bulat/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/create-pr
• npm test
• npm run validate

[aictrl-dev]

🟡 Eval result claims PASS but 2 of 9 criteria unaddressed.

🤖 Fix with your agent

Fix this code review finding (aictrl-dev/skills PR #27, evals/results.md:16-23):

Problem: Eval result claims PASS but 2 of 9 criteria unaddressed
Detail: The create-pr eval defines 9 pass criteria in evals/create-pr.eval.md, but the results table in evals/results.md has only 5 rows and the method narrative explicitly addresses only 7 of the 9. Two criteria are not evidenced anywhere in the result: (1) "Produces a concise title naming the endpoint behavior rather than files" (evals/create-pr.eval.md:22) and (2) "Keeps the linked issue close keyword when appropriate and preserves only accurate existing metadata" (evals/create-pr.eval.md:31-32) — the close-keyword/metadata criterion is entirely unmentioned despite the scenario containing a linked issue and stale metadata. The result records "Verdict: PASS" without demonstrating these. For a skill whose central thesis is that unverified claims must be marked unknown rather than asserted, recording PASS while leaving its own pass criteria unaddressed undercuts that thesis. Add table rows / narrative evidence for the title-naming and close-keyword/metadata criteria, or explicitly mark them not-applicable with reasoning.
Suggested fix: Add two rows to the results table, e.g. '| Title names endpoint behavior, not files | PASS |' and '| Linked-issue close keyword and existing metadata handled correctly | PASS |' (with the forward test actually evaluating these), or note in the method narrative why a criterion is out of scope.

Implement the fix on the PR head branch and add a regression test that fails before the fix and passes after.

Why this matters

The create-pr eval defines 9 pass criteria in evals/create-pr.eval.md, but the results table in evals/results.md has only 5 rows and the method narrative explicitly addresses only 7 of the 9. Two criteria are not evidenced anywhere in the result: (1) "Produces a concise title naming the endpoint behavior rather than files" (evals/create-pr.eval.md:22) and (2) "Keeps the linked issue close keyword when appropriate and preserves only accurate existing metadata" (evals/create-pr.eval.md:31-32) — the close-keyword/metadata criterion is entirely unmentioned despite the scenario containing a linked issue and stale metadata. The result records "Verdict: PASS" without demonstrating these. For a skill whose central thesis is that unverified claims must be marked unknown rather than asserted, recording PASS while leaving its own pass criteria unaddressed undercuts that thesis. Add table rows / narrative evidence for the title-naming and close-keyword/metadata criteria, or explicitly mark them not-applicable with reasoning.

| Criterion | Result |
|---|---|
| Pinned revision and unrelated working-tree boundary | PASS |
| Problem, behavior, scope, rollout/risk, and verification body structure | PASS |
| Authored rollout and test-result claims kept distinct from configuration and test source | PASS |
| Unknown deployment and execution evidence made explicit | PASS |
| Explicit confirmation required before host mutation | PASS |

Verdict: PASS — the skill produces a reviewer-ready draft that is grounded in
the exact change and honest about what has and has not been verified.

[byapparov]

TRUE · FIX — completed the create-pr eval record so it explicitly covers all nine pass criteria. The forward-test evidence now includes the behavior-focused title and the retained linked-issue close keyword/no unsupported metadata. Verified at ce5778b802d36af3227d7b1f872776dc0a0f159d with npm test (25 passed, 2 renderer-dependent skipped) and npm run validate (passed). Please re-review this focused update.

[aictrl-dev]

Code review

Verdict: Looks good — only minor / nit comments below. · 🔴 0 · 🟠 0 · 🟡 1 · ⚪ 0 · 0/1 resolved

• 🟡 evals/results.md:16-23 — Eval result claims PASS but 2 of 9 criteria unaddressed

🤖 Fix all 1 open findings with your agent

Fix the following code review findings on aictrl-dev/skills PR #27 (head branch).
Run the relevant tests/linters after each change.

1. evals/results.md:16-23 — Eval result claims PASS but 2 of 9 criteria unaddressed
   Detail: The create-pr eval defines 9 pass criteria in evals/create-pr.eval.md, but the results table in evals/results.md has only 5 rows and the method narrative explicitly addresses only 7 of the 9. Two criteria are not evidenced anywhere in the result: (1) "Produces a concise title naming the endpoint behavior rather than files" (evals/create-pr.eval.md:22) and (2) "Keeps the linked issue close keyword when appropriate and preserves only accurate existing metadata" (evals/create-pr.eval.md:31-32) — the close-keyword/metadata criterion is entirely unmentioned despite the scenario containing a linked issue and stale metadata. The result records "Verdict: PASS" without demonstrating these. For a skill whose central thesis is that unverified claims must be marked unknown rather than asserted, recording PASS while leaving its own pass criteria unaddressed undercuts that thesis. Add table rows / narrative evidence for the title-naming and close-keyword/metadata criteria, or explicitly mark them not-applicable with reasoning.
   Suggested fix: Add two rows to the results table, e.g. '| Title names endpoint behavior, not files | PASS |' and '| Linked-issue close keyword and existing metadata handled correctly | PASS |' (with the forward test actually evaluating these), or note in the method narrative why a criterion is out of scope.

📋 Out-of-diff findings (1)

Sev	Location	Finding
🟡	evals/results.md:16-23	Eval result claims PASS but 2 of 9 criteria unaddressed

Reviewed 7 files · 0 inline · view all 1 findings ↗

---

aictrl · AI code review for fast-moving teams · aictrl.dev

[aictrl-dev]

Code review

Verdict: Looks good. ✅ · 0 findings

Reviewed 7 files · 0 inline · view all 0 findings ↗

---

aictrl · AI code review for fast-moving teams · aictrl.dev

[aictrl-dev]

Code review

Verdict: Looks good — only minor / nit comments below. · 🔴 0 · 🟠 0 · 🟡 1 · ⚪ 0 · 0/1 resolved

• 🟡 skills/create-pr/SKILL.md:3 — Spec-vs-issue stance contradicts the mutation gate

🤖 Fix all 1 open findings with your agent

Fix the following code review findings on aictrl-dev/skills PR #27 (head branch).
Run the relevant tests/linters after each change.

1. skills/create-pr/SKILL.md:3 — Spec-vs-issue stance contradicts the mutation gate
   Detail: The skill sends mixed signals about whether a linked issue is required. The frontmatter description (line 3) and "Ground the draft" step 2 (line ~47) both list "linked issue or specification" as alternative grounding inputs, and the eval pass criterion says "Keeps the linked issue close keyword when appropriate"; yet the "Resolve the PR route and linked issue" section (line ~30) mandates a linked issue before any PR mutation: "Resolve the linked issue before preparing a mutation. ... do not create or edit the pull request until it is known." The skill never reconciles these positions, leaving behavior ambiguous for spec-grounded or no-issue PRs (dependency bumps, doc fixes, hotfixes). Either drop "or specification" from the description and step 2 to match the strict mutation gate, or add a clause describing how a spec-only request (or a user-confirmed no-issue PR) is handled at the mutation gate.
   Suggested fix: Reconcile the stance on linked issues: either (a) drop "or specification" from the description and from step 2 of "Ground the draft" so the strict mutation gate in "Resolve the PR route and linked issue" matches the rest of the skill, or (b) keep the spec alternative and add a short clause to the mutation gate describing how a verified spec (or an explicitly user-confirmed no-issue PR such as a deps bump or doc fix) is allowed to proceed.

📋 Out-of-diff findings (1)

Sev	Location	Finding
🟡	skills/create-pr/SKILL.md:3	Spec-vs-issue stance contradicts the mutation gate

Reviewed 20 files · 0 inline · view all 1 findings ↗

---

aictrl · AI code review for fast-moving teams · aictrl.dev

[aictrl-dev]

Code review

Verdict: Looks good. ✅ · 0 findings

Reviewed 20 files · 0 inline · view all 0 findings ↗

---

aictrl · AI code review for fast-moving teams · aictrl.dev