Qa risk review

BMAD Skill Mapping: QA Risk Review

Intent

Provide advisory QA and repair guidance around candidate artifacts while preserving EvalRunner as the canonical judge.

BMAD source/reference concept

Inspired by BMAD QA review, risk assessment, missing-test identification, and repair guidance practices.

CLike adaptation

Use this mapping in eval/qa to improve pre-eval hardening and advisory companion docs. It must never alter canonical eval verdicts.

Applies when

Applies to eval/qa.

Required inputs

• AGENT_EVAL_CONTEXT.json.
• Candidate source, tests, CI, LTC, HOWTO, and reports.
• SPEC, PLAN, plan.json, TECH_CONSTRAINTS, lane guides, and companion docs.

Required outputs

• Focused repair notes and QA advisory companion outputs under the target KIT root.
• Exact commands rerun and remaining blockers when repair is incomplete.

Companion outputs

• runs/kit/<REQ-ID>/docs/BMAD_QA_ADVISORY.md
• runs/kit/<REQ-ID>/docs/FIX_GUIDANCE.md
• runs/kit/<REQ-ID>/docs/MISSING_TESTS.md
• runs/kit/<REQ-ID>/docs/RISK_REVIEW.md
• runs/kit/<REQ-ID>/reports/BMAD_EVAL_REPAIR_NOTES.md

Downstream consumers

Canonical EvalRunner, Gate, KIT repair loops, and FINALIZE use the advisory output as context only.

Quality checks

• Root cause is tied to exact files, checks, and evidence.
• Repairs are candidate-owned and minimal.
• Missing tests and risks are named without hiding failures.
• Environment blockers include concrete evidence.

Eval/Gate evidence expectations

EvalRunner remains authoritative. Gate may consider advisory risk notes but must rely on CLike canonical reports and policy.

Forbidden behavior

• Do not decide pass/fail.
• Do not mutate canonical eval verdict fields.
• Do not modify canonical roots or dependency KIT roots.
• Do not weaken LTC, tests, typecheck, lint, security checks, or gate policy.
• Do not call BMAD runtime or external CLI.

Runtime dependency status

Reference-only. No BMAD QA runtime executes.

Cloud usage notes

Use concise advisory framing and preserve canonical eval authority.

Local-agent usage notes

Read selected skill context before repair. Patch only under allowed write roots and rerun the failing checks when possible.

Governance boundaries

CLike owns EvalRunner, Gate, write policy, reports, and promotion status.