Agent-Skills.md

Agent Skills: Self-Review Protocol

Perform a critical self-review of recently written code to catch bugs, edge cases, and quality issues before delivery. Use when user says "review your code", "check your work", "self-review", "fresh eyes", or after completing significant code changes.

UncategorizedID: l-lin/dotfiles/self-review

Repository

l-lin/dotfiles
l-lin
161

Install this agent skill to your local

pnpm dlx add-skill https://github.com/l-lin/dotfiles/tree/HEAD/home-manager/modules/share/ai/.config/ai/skills/self-review

Skill Files

Browse the full folder contents for self-review.

Download Skill

Loading file tree…

home-manager/modules/share/ai/.config/ai/skills/self-review/SKILL.md

Skill Metadata

Name
self-review
Description
Perform a critical self-review of recently written code to catch bugs, edge cases, and quality issues before delivery. Use when user says "review your code", "check your work", "self-review", "fresh eyes", or after completing significant code changes.

Self-Review Protocol

Purpose

After writing or modifying code, perform a thorough self-inspection to catch issues that are easy to miss during initial implementation.

Review Iterations

  • Default: Perform 3 independent review passes
  • User-specified: If user says "review N times", use that number
  • Each pass should approach the code with fresh perspective
  • Continue until all passes find no issues OR max iterations reached

Trigger Conditions

  • User explicitly requests review ("review your code", "check that", "self-review")
  • After completing substantial code changes (optional, use judgment)
  • Before marking a task as complete

Review Checklist

Examine ALL code written or modified in this session. Look for:

  1. Correctness

    • Obvious bugs, logical errors, or typos
    • Off-by-one errors, null checks, boundary conditions
    • Incorrect operators or comparison logic

  2. Edge Cases

    • Empty inputs, null values, undefined states
    • Boundary values (min/max, zero, negative)
    • Concurrent access issues
    • Resource exhaustion scenarios

  3. Error Handling

    • Missing try/catch blocks
    • Unhandled promise rejections
    • Silent failures
    • Poor error messages

  4. Code Quality

    • Dead code or unused parameters
    • Unnecessary complexity or duplication
    • Variables used before initialization
    • Type mismatches or unsafe casts
    • Resource leaks (file handles, connections)

  5. Consistency

    • Naming conventions
    • Code style alignment with project
    • Missing documentation for complex logic

Response Format

After each review pass:

If issues found:

  1. Fix them immediately
  2. Report: "Pass [X/N]: Fixed [M] issue(s)."
  3. List what was fixed (brief)
  4. Continue to next pass

If no issues found in current pass:

  1. Report: "Pass [X/N]: No issues found."
  2. List what was verified (brief)
  3. Continue to next pass

After all passes complete:

  • Report total: "Completed [N] passes. Total fixes: [M]. Review complete."

Rules

  • Think deeply: Don't rush to a verdict
  • Be skeptical: Assume there might be issues
  • Be thorough: Check every function, every branch
  • Be honest: Report issues even if embarrassing
  • No defensiveness: If you find a bug, just fix it
  • Independent passes: Each review pass should use a different mental model or focus area
  • No shortcuts: Complete all N passes even if early passes find nothing

Example Multi-Pass Response

Pass 1/3: Fixed 2 issue(s).
1. Added null check in `parseInput()` - crashed on empty string
2. Fixed off-by-one in loop boundary - skipped last element

Pass 2/3: Fixed 1 issue(s).
1. Removed unused parameter `options` in `validate()`

Pass 3/3: No issues found.
Verified:
- All 4 functions handle null/undefined inputs
- Error paths tested with edge cases (empty, zero, negative)
- No unused variables or dead code
- Resource cleanup in finally blocks
- Type safety maintained throughout

Completed 3 passes. Total fixes: 3. Review complete.