code-reviewer
Comprehensive review agent that verifies spec compliance, logic correctness, and API contracts with severity-rated findings.
The code-reviewer agent is the primary gatekeeper before changes reach production. It runs a strict two-stage review: spec compliance first, then code quality. Every finding is rated by severity (CRITICAL, HIGH, MEDIUM, LOW) and includes a concrete fix suggestion. The agent never approves code with CRITICAL or HIGH severity issues and never skips to style nitpicks before confirming the implementation actually satisfies the request.
Role
- Verify that the implementation covers all requirements and solves the right problem (Stage 1)
- Run
lsp_diagnosticson every modified file and flag type errors before issuing a verdict - Rate each finding by severity and provide a specific, actionable fix suggestion (Stage 2)
- Issue a clear verdict: APPROVE, REQUEST CHANGES, or COMMENT
When invoked
- By
$code-reviewwhen a developer requests a comprehensive review before merging - By
$autopilotat designated review checkpoints after an executor completes a task - On pull requests that touch API contracts, authentication, or data handling
- When a
verifierpass surfaces correctness questions that require deeper inspection
Inputs
- Git diff of the changes under review
- Spec, PR description, or issue description to confirm intent
- Full file context around modified sections, read via Grep and Read
- LSP diagnostics output for all changed files
Outputs
- A review report listing findings by severity with
file:linereferences and fix suggestions - A count of issues per severity tier (CRITICAL / HIGH / MEDIUM / LOW)
- A final verdict (APPROVE / REQUEST CHANGES / COMMENT) based on the highest severity found
Limits
- Read-only — Write and Edit tools are not available during review
- Does not approve any change with a CRITICAL or HIGH severity finding outstanding
- Does not skip Stage 1 (spec compliance) to prioritize style feedback
Related agents
- quality-reviewer — focused deep-dive on logic defects and maintainability
- security-reviewer — dedicated OWASP and secrets scan when security depth is needed
- style-reviewer — lightweight formatting and naming pass
- api-reviewer — specialized backward-compatibility and versioning analysis
explore-harness
Evidence-driven tracer that manages competing hypotheses with for/against evidence, uncertainty scores, and next-probe recommendations.
security-reviewer
Security vulnerability detection agent covering OWASP Top 10, secrets exposure, trust boundaries, and authentication/authorization flows.