sunpar

Use when executing a slice plan wave by wave.

Use when validating generated slices.

Use after PR creation when requesting Codex or Copilot review.

Author acceptance tests from requirements before implementation.

Evaluate plans/tasks/outputs for schema and quality violations.

Audit API/schema/route/CLI contract changes for compatibility and migration needs.

Stage changes, commit, push, and create/update PR with review-ready hygiene.

Validate dependency additions/upgrades for risk and security.

Commit and publish local GitHub work with review requests. Use when the user wants Codex to commit staged changes, stage all changes only when nothing is staged, push the current branch, and open a GitHub pull request that tags @codex and @copilot for review.

Analyze staged and unstaged git changes and refactor for readability, maintainability, and conciseness while preserving behavior. Use when the user asks to refactor recent changes, clean up work in progress, improve code quality, or review a diff for safe structural improvements.

Generate implementation TASKS.md plans from one design document or from all feature documents in a directory.

Run a phased feature-development workflow: discovery, codebase exploration, clarifying questions, architecture options, implementation, quality review, and summary. Use when a user asks for end-to-end feature delivery with strong design rigor before coding.

Convert one planning document into PRD.md, USER_FLOW.md, BACKEND_ARCHITECTURE.md, FRONTEND_DESIGN.md, and TASKS.md in fixed deterministic order with a quality gate and traceability checks.

Remove AI-generated code slop from the current branch by auditing `git diff main...HEAD` and applying minimal cleanups. Use when a user asks to de-slop a PR/branch, make code look human-written, or remove over-defensive/verbose/generated patterns without changing behavior.

Execute a full feature workflow in order by chaining feature development, test writing, feature summaries, refactor analysis, selective refactor implementation, personal code review artifacts, and review-feedback triage when feedback is available. Use when the user wants an end-to-end structured delivery process.

Triage code review feedback from chat text or a markdown file, validate each item against the codebase, produce a fix plan, and delegate implementation to subagents. Use when the user wants every review comment verified before applying fixes.