team-verification-guard

name	team-verification-guard
description	Use when verifying completed work, checking OpenSpec readiness, running tests, reviewing evidence, or deciding if ready to archive.

Team Verification Guard

Core Principle

Verification must be layered. Each layer must pass before proceeding to the next. No evidence = no claim of readiness.

When to Use

Use this skill when:

• Executing /team-verify layered verification
• Deciding whether to archive
• Checking task completion
• Verifying acceptance criteria alignment

Do NOT use this skill when:

• Creating proposals (use team-openspec-guard)
• Implementing code (use team-implementation-guard)

Layer 1: OpenSpec Validity

openspec validate <change-id> --strict

Checks:

• Spec delta format correct (## ADDED/MODIFIED/REMOVED/RENAMED Requirements)
• Every requirement contains SHALL/MUST
• Every requirement has ≥ 1 #### Scenario:
• No conflicting sections (e.g., both ADDED and REMOVED)
• No zero errors

Layer 2: Task Completion

Check tasks.md:

• Every required task marked - [x]
• Every completed task has evidence
• Deferred tasks have reasons and follow-up plan

Task	Status	Evidence
1.1	✅	pytest tests/models/test_user.py — 8 passed
1.2	✅	Manual: http://localhost:3000/ verified
2.1	❌	Blocked: API contract undecided

Layer 3: Project Checks

Check	Command	Result
Lint	npm run lint	0 errors
Unit Tests	pytest tests/	all passed
Build	npm run build	success
Typecheck (if TS)	tsc --noEmit	0 errors

Layer 4: Acceptance Criteria

AC	Evidence	Status
AC-1: ...	pytest tests/...	✅
AC-2: ...	Manual: ...	⚠

Non-Negotiable Rules

• Do NOT claim "ready" if any required task is incomplete
• Do NOT claim "ready" if openspec validate --strict fails
• Do NOT claim "ready" if you have not run tests (or explained why not)
• Do NOT infer "passing" from the absence of errors in chat — run the actual commands
• Do NOT hide skipped checks — list them explicitly with reasons
• Do NOT mark an AC as passed without evidence

Layered Checklist

Layer 1:

• openspec validate <change-id> --strict passes
• No zero errors
• Warnings reviewed and confirmed non-blocking

Layer 2:

• X / Y tasks completed
• Each completed task has evidence
• Deferred tasks: reason + follow-up plan

Layer 3:

• Lint: 0 errors
• Unit tests: all pass
• Build: success
• Typecheck (if applicable): 0 errors

Layer 4:

• Every AC has matching evidence
• Evidence is verifiable (not "looks fine")
• Failures/missing items marked and recorded

Archive Readiness

Must satisfy:

• All required tasks marked - [x]
• openspec validate <change-id> --strict passes
• All ACs have evidence
• All project checks pass (lint / test / build)
• No unresolved hard-blocking open questions

Conditional (archive with notes):

• Deferred tasks: reason + follow-up plan
• Skipped checks: alternative verification path
• Risks: assessed and recorded

Cannot archive:

• OpenSpec validation fails
• Required tasks incomplete
• Core ACs lack evidence
• Project checks not run
• Unresolved hard blockers exist

Final Verdict

• ✅ Ready: Layers 1-4 all pass
• ⚠ Conditional: Critical risks exist but documented
• ❌ Not Ready: Any layer has blocking issues

Common Mistakes

Mistake	Fix
Claiming "looks fine"	Run specific commands and provide output
Running only partial tests	Run the full test suite
Skipping lint/build	At minimum confirm core checks pass
Marking AC passed without evidence	Attach test name or manual steps to each
Hiding skipped checks	List skipped items and reasons explicitly

Bad Cases

When this skill fails, record in badCases/ directory:

• Input: what the user said
• Wrong output: what the AI did that it shouldn't have
• Expected output: what it should have done
• New rule needed: how to prevent recurrence

See badCases/ for case history.