Review code
Review all uncommitted changes (the Test phase) in a SEPARATE verifier — never self-review. Use before securing/committing.From its SKILL.md
npx -y skills add NilsWidal/loobster --skill review-codeAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
SKILL.md
3.0 KB, as published. Nobody here has run it
Perform a comprehensive review of all uncommitted changes.
Independent review — never self-verify. This review MUST be run by a separate verifier subagent (
Agent) that did not write the code. The implementing agent does not review its own work. Spawn a judge with the diff + what to check; it returns only findings. (See the "Never self-verify" rule inrun.md.) Apply theverifymodel preference from.claude/loobster.jsonwhen spawning the judge (see run.md "Preferred subagents"); no config → inherit the session model.
What this does
- Gathers and reviews all uncommitted changes in the current branch
- Produces a prioritized list of action items with file:line references
- Optionally posts findings to the relevant issue tracker
Steps
- Collect change context:
git status --porcelaingit diffgit diff --cachedgit diff HEADgit log --oneline -n 5
- If an issue ID is provided or can be inferred from the branch name, read the issue to understand the intent.
- Analyze changes for: security, performance, style, consistency, missing edge cases, dependency impacts, and integration risks.
- Token discipline (see
.agents/skills/token-discipline/SKILL.md): for a wide diff, delegate per-area review toAgentsubagents (e.g. one per dimension or per large file). Each subagent receives its slice of the diff plus what to check, and returns only a findings list (each: severity +file:line+ one-line fix) — never the raw hunks. The main thread merges those findings into the single prioritized action list below without re-reading the hunks the subagents already saw.
- Token discipline (see
- If the diff touches frontend (
.tsx/.jsx/.vue/.svelte/.html/.css/.scssor component/route dirs), run the verifiable frontend layer — follow.agents/skills/verify-frontend/SKILL.mdto capture Playwright screenshots and attach them to the PR. A render failure (4xx/5xx or console errors) is a must-fix. (Independent: CI or a separate verifier subagent assesses — not the implementer.) - Output a summary and a single prioritized action list using indicators:
- RED — must-fix
- YELLOW — recommended
- GREEN — consider
- If Linear MCP tools are available and an issue was identified, post the review as a comment on the issue.
- If there are no action items, suggest the issue is ready for the next phase.
- If Claude Code Tasks track this work (
TaskList), reflect the outcome: leave any task with open action items asin_progress, and only mark taskscompletedonce their action items are resolved. This keeps/resumeaccurate.
Output template
## Code Review
Summary: <1-2 sentences>
Action Items:
1. <indicator> <action> in `path:line`
2. <indicator> <action> in `path:start-end`