verification-protocol

name	verification-protocol
description	Canonical verification protocol for Sando Design System agents. Provides the exact commands, decision trees, and thresholds each specialist agent must run after completing work. Replaces per-agent `<verification>` blocks that are duplicated with slight variations across agents. Trigger: inject into sub-agent prompts so each agent knows exactly what to run and what "done" means.
license	MIT
metadata	{"version":"1.0.0","author":"Sando Design System"}

Verification Protocol

Canonical post-work verification for every Sando specialist agent. Inject the matching section into sub-agent prompts under ## Verification Protocol (auto-resolved).

---

DEVELOPER Verification

Run in sequence after any code change:

pnpm lint        # ESLint — must pass (0 errors)
pnpm typecheck   # TypeScript — must pass (0 errors)
pnpm test        # Vitest — must pass (all green)

Decision tree:

lint fails    → fix lint errors → re-run lint
typecheck fails → fix type errors → re-run typecheck
tests fail    → fix tests (or update if behavior intentionally changed) → re-run tests

ALL PASS → mark complete
ANY FAIL → mark partial/blocked — never mark complete

Quick path (small non-logic changes): pnpm lint && pnpm typecheck

---

TOKENS Verification

Run after any token file change:

pnpm tokens:build          # Style Dictionary — must succeed
ls packages/tokens/dist/css/  # Verify output exists

Decision tree:

build fails → check JSON syntax → verify all references exist → fix → rebuild
output missing → re-run build → check config

BUILD PASSES + output present → mark complete
BUILD FAILS → mark partial — never mark complete

---

QUALITY Verification

Run after writing or modifying tests:

pnpm test              # All tests must pass
pnpm test:coverage     # Coverage report

Thresholds (blocking):

Metric	Threshold
Unit test coverage	≥ 80%
A11y test coverage	100% for public components
axe-core violations	0 critical or serious

Decision tree:

tests fail       → fix test or report bug in component → re-run tests
coverage < 80%   → add more tests → re-run coverage
a11y violations  → report to sando-developer for fix → re-test

ALL THRESHOLDS MET → mark complete
ANY THRESHOLD MISSED → mark partial — never mark complete

---

STORYBOOK Verification

After creating or modifying stories:

pnpm docs:build        # Must succeed (0 errors)

Manual checklist:

• All story controls render correctly
• FlavorDecorator applies without errors
• No console errors in browser
• A11y addon shows 0 violations per story

Decision tree:

build fails   → fix story syntax/imports → rebuild
a11y failures → report to sando-developer or sando-quality

BUILD PASSES → mark complete
BUILD FAILS  → mark partial

---

ARCHITECT Verification

After proposing a design or ADR:

pnpm build     # Full monorepo build — must still pass

Checklist:

• Decision documented in ADR format (title, status, context, decision, consequences)
• Token layer references are correct (no layer violations)
• Breaking changes flagged with migration guide
• Affected .toon guideline files updated or flagged for update

---

DOCUMENTER Verification

After writing docs or JSDoc:

pnpm typecheck   # JSDoc types must be valid

Checklist:

• All public @property decorators have JSDoc
• @fires documented for all custom events
• @slot documented for all named slots
• @csspart documented for all CSS parts
• No broken links in Storybook MDX docs (?path=/docs/<id>--docs)

---

Compact Rules (for skill-registry)

• Inject the matching agent-role section into sub-agent prompts under ## Verification Protocol (auto-resolved)
• Agents MUST run the specified commands before marking STATUS: complete
• Never mark complete if any blocking check fails — use partial or blocked
• The Return Envelope STATUS field maps directly to these outcomes:
  • complete = all verification checks passed
  • partial = some deliverables done but verification failed
  • blocked = cannot proceed (missing dependency, critical build error)