Code Review

Two-pass self-reflection scoring review pipeline.

Code Review

A two-pass self-reflection scoring pipeline:

  1. Generate: find potential issues using universal and technology-specific rules.
  2. Score: re-evaluate each finding with a confidence of 1 to 10.
  3. Filter: drop noise below the mode threshold.
  4. Report: severity-grouped output with fix suggestions.

Technology contexts are auto-detected for 14 stacks, including Laravel, Vue/Nuxt, Angular, Go, and Rust.

Commands and modes

  • /code-review:code-review runs a single-agent review with --mode (quick, thorough, security, pr) and --threshold.
  • /code-review:team-review spawns parallel reviewers for security, performance, and correctness.
ModeThresholdUse case
Quick>= 7Small changes, fast feedback
Thorough>= 4Large changes, quality gate
Security>= 3 security / >= 7 otherSecurity audits
PR>= 5Pull request with verdict