uu-review

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# UU Review

Review the current changeset against its canonical `docs/uu-<task-slug>.md` plan. Treat pasted worker summaries and user notes as leads, not evidence. A pasted `uu-second-opinion` report is a high-value review input, not direct implementation direction.

## Rules

- Read repository instructions, the canonical plan, and the actual change surface. Do not edit or write. Never stage, commit, or push; only the human may do so.
- Identify the changeset from context, status, diffs, untracked files, and history. If necessary, ask for the plan path or commit/range.
- Verify material worker claims and plan outcomes in code and tests. Prefer focused executable checks to assertions.
- Review only the intended changeset and necessary surrounding code. Report actionable issues, not preferences.
- Investigate every substantive `uu-second-opinion` finding. Retain it unless the code, canonical plan, tests, or other concrete repository evidence shows it is incorrect, already resolved, or not a real defect. Do not reject it merely as out of scope; give specific evidence for every rejection.
- When a second-opinion report includes lower-priority observations with a P0/P1 finding, adjudicate them in that same review loop. They do not independently reopen work.

## Check

Look for missing plan outcomes, incorrect behavior, regressions, compatibility failures, unsafe edge cases, security or data-loss risks, concurrency defects, and missing meaningful tests. Where practical, ensure a regression test would fail without the fix; otherwise, state why and describe the strongest available validation. Use broader checks only when the change warrants them.

Before reporting, reset the review: re-audit the complete changed boundary after fix rounds; treat prior findings and passing tests as leads, not proof; independently validate affected integrations and meaningful edge cases.

## Report

Lead with `APPROVE`, `APPROVE WITH MINOR ISSUES`, `REQUEST CHANGES`, or `BLOCKED`. Group related findings by priority:

- **P0:** catastrophic risk
- **P1:** blocking correctness or compatibility failure
- **P2:** defect or important missing coverage
- **P3:** limited-impact follow-up

For each finding, give evidence, failure scenario, impact, and correction. Then briefly state the plan/summary claim audit, checks actually run and outcomes, and residual risks. Use `REQUEST CHANGES` for P0–P2; use `APPROVE WITH MINOR ISSUES` only for P3.

From searlelab/until-useful, Apache-2.0.

More from searlelab/until-useful