adlc-review - SKILL.md Agent Skill

name: adlc-review description: Review staged or working-tree changes with ADLC context and read-only gate output.

ADLC Review

Use this for correctness, regression, maintainability, and risk review.

Process

Resolve effective ADLC paths from .adlc/config.yaml; use adlc resolve-config when the CLI is available.
Pin the diff target.
Read the active plan, configured rules, and architecture.
Review behavior first, then maintainability, security, docs, and release risk.
Prefer review-sidecar, security-sidecar, and rules-sidecar for independent read-only checks when available.
Classify blockers as autonomous-fixable or human-gated. Code, docs, tests, rules, and security findings should suggest adlc-fix; human-gated blockers are limited to explicit decisions, credentials or external accounts, destructive or production operations, legal/security sign-off, scope or product ambiguity, or user-requested approval points.
Lead with blocking findings. Say clearly when there are none.

Output

End with a final parseable adlc-gate-result fenced block:

{
  "schema_version": 1,
  "gate": "review",
  "status": "pass|warn|fail",
  "blocking": false,
  "blockers": [],
  "affected_files": [],
  "suggested_next": {
    "command": "adlc-commit|adlc-fix|null",
    "reason": "Short reason."
  }
}

Use suggested_next.command: "adlc-fix" for autonomous-fixable blockers. Use null only when a human-gated blocker prevents the next autonomous gate.