abildtoft/kramme:pr:autoreview

Auto Review

Run the local structured PR code review as a closeout check, then verify and handle any actionable findings. This skill adapts the OpenClaw autoreview workflow to the Kramme review stack by delegating the review itself to kramme:pr:code-review.

Use When

• The user asks for autoreview, a final code-review pass, or second-model review.
• Non-trivial code edits are ready for closeout before final response, commit, push, or PR creation.
• A local branch or PR branch needs another review pass after fixes.

Do not use this for UX, visual, accessibility, or product review. Use kramme:pr:ux-review, kramme:visual:*, or kramme:pr:product-review instead.

Contract

• Treat review output as advisory. Never apply a finding blindly.
• Verify every accepted finding against the real code path and adjacent files before changing code.
• Reject speculative risks, unrealistic edge cases, broad rewrites, and fixes that do not fit the local codebase.
• Prefer the smallest fix at the correct ownership boundary.
• If an accepted finding reveals a repeated bug class, inspect the current review scope for sibling instances before fixing.
• If a review-triggered fix changes code, rerun focused verification and rerun the review.
• Stop when the rerun has no accepted/actionable Critical or Important findings, or when remaining findings are clearly manual/advisory and reported as such.
• Do not push changes unless the user explicitly asked for push, ship, or PR update.

Workflow

1. Prepare arguments

   • Treat $ARGUMENTS as arguments for kramme:pr:code-review.
   • Pass through supported review arguments such as aspect names, --base <branch>, --inline, --parallel, parallel, and --emphasize <dimension>.
   • If no arguments are provided, run the default full kramme:pr:code-review workflow.

2. Run the delegated review

   • Invoke /kramme:pr:code-review $ARGUMENTS.
   • If slash-skill invocation is unavailable, locate and read the kramme:pr:code-review skill definition in the installed skills directory and follow that workflow with the same arguments.
   • Do not duplicate the reviewer taxonomy, diff collection, relevance validation, or report formatting here; those belong to kramme:pr:code-review.

3. Read the review result

   • If --inline was used, inspect the inline review response.
   • Otherwise read REVIEW_OVERVIEW.md from the project root.
   • If the delegated review reports degraded coverage or failed reviewers, tell the user which dimensions were not covered before acting on findings.

4. Triage findings

   • For each Critical or Important finding, verify the evidence in the code before accepting it.
   • Reject findings that are pre-existing, out of scope, already addressed, unverified without a concrete failure path, or inconsistent with local patterns.
   • Keep manual findings as manual unless the missing context can be resolved locally without guessing.
   • Treat Suggestions and FYI as optional unless the user asked for a stricter cleanup pass.

5. Resolve accepted findings

   • For verified gated_auto Critical or Important findings, use /kramme:pr:resolve-review --severity critical,important when it fits the report's recommended action. The severity filter keeps this loop from processing suggestion-severity advisory findings it never triaged.
   • If --inline was used, pass the inline review content to /kramme:pr:resolve-review or make the smallest scoped fix directly; do not rely on local review-file discovery.
   • For verified findings that need hand edits, make the smallest scoped fix directly.
   • Do not resolve manual findings without enough product, ownership, or reviewer context.

6. Verify and rerun

   • Run focused tests, type checks, lint, or build commands that cover the code changed while resolving review findings.
   • Rerun /kramme:pr:code-review $ARGUMENTS after review-triggered code changes.
   • Continue until there are no accepted/actionable Critical or Important findings, or report the remaining manual/advisory items clearly.

Artifact Lifecycle

• kramme:pr:code-review produces REVIEW_OVERVIEW.md by default unless --inline is used.
• kramme:pr:resolve-review consumes REVIEW_OVERVIEW.md when resolving eligible file-backed findings, or explicit inline review content when --inline was used.
• Rerunning this skill refreshes the delegated review report.
• /kramme:workflow-artifacts:cleanup retires REVIEW_OVERVIEW.md when the review artifact is no longer needed.

Examples

/kramme:pr:autoreview
/kramme:pr:autoreview --parallel
/kramme:pr:autoreview tests errors --base origin/main
/kramme:pr:autoreview --inline --emphasize security