code-review-policy

name: code-review-policy description: "Passive policy for LLM code review findings, severity taxonomy, and evidence format." user-invocable: false

Code Review Policy

Scope

• RTL logic correctness and spec/uarch compliance
• Testbench quality and checker/scoreboard adequacy
• Verification completeness (coverage + traceability)
• Silicon risk indicators (constraints/CDC/synthesis/timing impacts)

Severity

• S0 Blocker: data corruption, protocol/safety breakage, hard spec violation
• S1 Critical: high silicon risk, functional failure likely under valid stimulus
• S2 Major: maintainability/test-quality issues with medium defect risk
• S3 Minor: style/readability/non-blocking improvements

Pass/Fail Gate

• FAIL: any unresolved S0 or S1
• CONDITIONAL: only S2 findings remain and have approved remediation plan
• PASS: only S3 or no findings

Escalation

• Any S0: immediate stop, escalate to orchestrator/user with fix-first recommendation
• Repeated S1 on same module after 2 fix loops: escalate to architecture owner
• Missing evidence for S0/S1: finding is invalid until evidence attached

Evidence Rules

• Every S0/S1 finding must include reproducible evidence:
  • file and line
  • failing command/log snippet
  • expected vs actual behavior
• Prefer replay artifacts (*_latest.sh, wave/report paths) over prose-only claims

Output Format

Use the following report skeleton:

# Code Review Report
- Scope: module|block|top
- Verdict: PASS | CONDITIONAL | FAIL

## Findings
| ID | Severity | File:Line | Summary | Evidence |
|---|---|---|---|---|

## Required Actions
| Priority | Owner | Action | Recheck |
|---|---|---|---|

## Escalations
- [if any]