rtl-p4s-bugfix - SKILL.md Agent Skill

name: rtl-p4s-bugfix description: "RTL bug fix workflow enforcing the full cycle: analyze → fix → lint → TB create/update → functional verification. Prevents RTL changes from being considered complete with lint-only validation." user-invocable: true

A workflow skill that enforces the design-to-verification flow when fixing RTL bugs. A fix is only considered complete after TB generation and functional simulation pass — not just lint.

Core principle: lint is only a syntax check, not evidence of functional correctness.

This skill integrates with the PostToolUse:Edit hook tracking system. Modified .sv files are automatically tracked, and session termination is blocked without functional verification.

- Fixing RTL bugs found during Phase 4 review - Fixing functional errors in RTL modules - Adding or modifying functionality in existing RTL - Verifying no functional regression after refactoring - Fixing integration bugs spanning multiple modules - **Phase 5→4 Feedback Loop**: Automatic fix path invoked when Phase 5 verification FAILs - Coding convention changes only (e.g., port renaming with no functional change) → use rtl-p4s-refactor - Writing a new module from scratch → use rtl-p4-implement - Simple lint error fixes (e.g., removing unused signals) → use rtl-lint-check Born from a previous session where 9 RTL bugs were fixed with only lint runs, skipping TB/simulation entirely. Passing lint is merely "compilation success" — simulation is required to prove the fix is functionally correct.

Anti-pattern example:

312 lines of RTL modified across 5 Waves → only verilator --lint-only executed → 0 TBs, 0 simulations
Result: declared "complete" with zero functional correctness verification

The orchestrator handles analysis, fix+lint, TB creation/update, functional verification, Phase 5→4 feedback return, parallel UNIT_FIX across modules, and lesson-learned recording.

All policies, checklists, escalation rules, and parallel fix decision logic are defined in the rtl-p4s-bugfix-policy skill (loaded via the orchestrator's skills: field).

Execution

Task(subagent_type="rtl-agent-team:p4s-bugfix-orchestrator", prompt="Execute RTL bug fix cycle. Bug description: $ARGUMENTS")

Do not perform any work directly. The orchestrator agent manages the full analyze → fix → lint → TB → simulation cycle.

Output

Fixed RTL file(s) with lint re-pass confirmation
Updated or new testbench with passing functional simulation results
.rat/scratch/phase-4/bugfix-decision-{N}.md — decision record documenting root cause, fix rationale, and verification evidence