AgentSkillsCN

self_review

对当前代码变更进行简洁而优先级分明的审查，提出切实可行的修复方案，重点关注 Torus（尤其是 Elixir/Phoenix 规范）中的正确性、可靠性、性能与可维护性，同时保持输出精炼且易于行动。

SKILL.md
--- frontmatter
name: self_review
description: >
  Perform a concise, prioritized review of current code changes with concrete fixes, focusing on correctness, reliability, performance, and maintainability in Torus (especially Elixir/Phoenix conventions) while keeping output compact and actionable.
examples:
  - "$self_review"
  - "Review this diff for risks before I open a PR ($self_review)"
  - "Quick self review of changed files, keep it concise"
when_to_use:
  - "Any implementation is complete and needs a quality pass."
  - "User asks for review, risk scan, or PR readiness check."
  - "A skill (for example $spec_develop) requires a final self-review gate."
when_not_to_use:
  - "No meaningful changes are present to review."
  - "The user wants a full rewrite instead of issue-oriented review."
  - "The task is pure product/spec drafting with no code diff."

Purpose

Return a short, prioritized list of issues and suggested fixes for current changes.

Inputs

  • •Current diff / changed files
  • •Optional context docs (prd.md, fdd.md, plan.md) when available
  • •Review guides (required for every self review):
    • •.review/elixir.md
    • •.review/performance.md
    • •.review/requirements.md
    • •.review/security.md
    • •.review/typescript.md
    • •.review/ui.md

Outputs (files changed/created)

  • •No required files
  • •Optional: small follow-up fixes if requested or required by calling workflow

Process (step-by-step)

  1. •Inspect changed files and identify behavior-impacting risks first.
  2. •Read all six .review/*.md guides and run each review/checklist against the current changes:
    • •Elixir review: .review/elixir.md
    • •Performance review: .review/performance.md
    • •Requirements review: .review/requirements.md
    • •Security review: .review/security.md
    • •TypeScript review: .review/typescript.md
    • •UI review: .review/ui.md
  3. •Review against this consolidated checklist:
    • •Correctness and edge cases
    • •Error handling and failure paths
    • •Elixir patterns (pattern matching, with, clear return contracts)
    • •Supervision/process boundaries when concurrency is involved
    • •Telemetry/logging hygiene (signal quality, no noisy or sensitive logs)
    • •Test quality and regression coverage
    • •Performance footguns (N+1, expensive loops, unbounded work)
    • •Readability and maintainability
  4. •Prioritize findings by severity: high, medium, low.
  5. •Provide concrete suggestions for each finding (BUT DO NOT MAKE THESE FIXES)

Quality bar / guardrails

  • •Keep output concise: maximum 20 bullets total.
  • •Findings first, no long summary before issues.
  • •Avoid speculative nits; focus on real risk and behavior impact.
  • •The six .review/*.md guides are mandatory inputs on every run.
  • •Review comments from the six guide-based reviews must be addressed before completion.
  • •If no findings, say so explicitly and mention residual test gaps if any.

Final response format (brief)

  • •Findings: bullets ordered high -> medium -> low
  • •Residual risks/test gaps: short bullets (or None)