Review
Claude Code plugin and marketplace of skills & commands for the full dev loop: /task → /plan → /do → /review → /gca → /gp → /pr. Plus /prd, /issues, /explore, /grill, /bootstrap, /handoff and more.
npx -y skills add yokeloop/yoke --skill reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Finds problems in code, fixes them and produces a report. Used when the user writes "review", "code review", "find issues", "find bugs", "check code", "code analysis", "prepare a report", "what's wrong with the code".
SKILL.md
2.2 KB, as published. Nobody here has run it
Review
Review the code for a task, fix the problems worth fixing, and leave behind a report a human reviewer can act on. A standalone utility — run it whenever code needs a second pass, not only inside a larger flow.
The orchestrator drives the run: it talks to the user, dispatches sub-agents for analysis and for fixes, and pauses only twice — to choose which findings to fix, and to choose what happens next. Every other step runs to completion without asking.
Input
$ARGUMENTS — a task-slug or a path to a task-file. With no argument, determine the slug from the current branch or the latest .yoke/ai/*/ directory. This input contract is stable, so callers that pass a slug keep working.
Procedure
The full pipeline — scope resolution, the code-reviewer dispatch, finding classification, the fix loop, validation, and report assembly — lives in reference/review-procedure.md. Read that file and execute it end to end. The report follows the template in reference/review-format.md.
Rules
- Standalone utility. Review runs on its own; it never assumes a larger flow around it.
- Fix, then report. Findings the user accepts are fixed and committed; everything found — fixed or skipped — lands in the report.
- Report location. Write the report to
.yoke/ai/<slug>/<slug>-review.md. - Every fork is one question. For any decision — fix scope, next action — ask a single AskUserQuestion with the recommended option listed first.
- Delegate the work. File edits, bash, and analysis go to sub-agents; each receives only its own data.
- Read the git memory. Pull decision context from the affected files' commit history per
${CLAUDE_PLUGIN_ROOT}/skills/gca/reference/history-reading.md— a finding that contradicts aConstraint:or re-proposes aRejected:approach must cite the commit it argues with. - Language: match the ticket/input language, or follow the project-level definition in CLAUDE.md / AGENTS.md.