Skip to content

verification.md

Latest commit

 

History

History
77 lines (49 loc) · 2.11 KB

verification.md

File metadata and controls

77 lines (49 loc) · 2.11 KB
Raw

Agent verification chain

Mandatory sequence for every agent after code changes — implementation, bugfix, refactor, or post-review fixes.

Agents must run these themselves. Do not ask the user to run them. Do not mark work complete until all steps pass.

Command reference: .claude/commands/smell.md · /smell [branch] (default base: develop)

When to run

Phase Steps
After implementation / fix Fast lint → /smell → fix BLOCKER/HIGH → make audit
Code review (PR or diff) /smell first → review-agent checklists
Review → fix cycle Apply fixes → repeat full chain below
Before PR or “done” Full chain; zero BLOCKER/HIGH smell findings

Chain (in order)

1. Fast lint (touched paths)

Go:

golangci-lint run ./cmd/... ./internal/... ./pkg/...
go test -race ./cmd/... ./internal/... ./pkg/...

Frontend:

cd frontend && npm run lint && npm run test

Fix compile/lint failures before smell — smell is not a substitute for lint.

2. /smell develop

Follow .claude/commands/smell.md on the git diff vs develop.

  • Produce the Smell Report (includes Principles checked: Clean Code · KISS · DRY · DDD · SOLID)
  • Fix all BLOCKER and HIGH catalog findings (MGW.*, CC.*, DDD.*, SOLID.*, GO.*, TS.*)
  • Re-run /smell after fixes until clean or only MEDIUM/LOW/NIT remain

3. make audit

From repository root:

make audit

Exit code 0 required. Fix failures and repeat from step 2 if the diff changed materially.

4. Report to user

State explicitly:

  • Smell summary (findings count / clean)
  • make audit result
  • Any MEDIUM items deferred (with reason)

Specialist routing

Agent Smell role
master-agent Enforces this chain on all delegated work
delivery-agent Runs chain after §4.3 Implement
review-agent Runs /smell before §2 checklists

See prompts.md for copy-paste snippets.