renatocaliari/skills/local/cali-testing

Testing Protocol

After implementing any feature, run this protocol before marking complete.

Contents

| Phase | What | When to skip | |-------|------|--------------| | Phase 1: Unit Tests | Run test suite, block on failure | Never | | Phase 2: Code Review | Parallel subagent review | <3 files changed | | Phase 3: UI Quality | Accessibility + design audit | Non-visual features | | Phase 4: Browser Testing | Interactive QA | Non-interactive features | | Phase 5: Final Checklist | Pre-completion verification | Never |

Phase 1: Unit Tests

Run the project's test suite first:

# Go
go test ./...

# Node
npm test

# Python
pytest

Block until tests pass. Do not proceed with failing tests.

Phase 2: Parallel Code Review via Subagents

Launch fresh-context reviewers in parallel:

subagent({
  tasks: [
    {
      agent: "reviewer",
      task: "Review this diff for correctness, regressions, and edge cases. Focus on: logic errors, missing error handling, security issues, performance regressions. Provide specific line references.",
      output: false
    },
    {
      agent: "reviewer",
      task: "Review this diff for simplicity and code quality. Focus on: unnecessary complexity, dead code, naming clarity, adherence to project conventions. Remove slop and verbosity.",
      output: false
    }
  ],
  concurrency: 2,
  context: "fresh"
})

When to use subagents:

• Diff touches 3+ files
• Feature involves multiple components
• Changes affect critical paths (auth, payments, data)

When to skip:

• Single-file typo fix
• Config-only change
• Documentation update

Phase 3: UI Quality (if visual)

Only if the scope involves a visual interface.

Accessibility Audit

Load the audit skill for WCAG compliance:

/audit

Checks:

• Color contrast ratios
• Keyboard navigation
• Screen reader compatibility
• ARIA attributes
• Focus management

Design Review

Load the critique skill for design quality:

/critique

Checks:

• Cognitive load
• Visual hierarchy
• Consistency with existing patterns
• AI slop detection (over-generated UI)

Phase 4: Browser Testing (if interactive)

Load agent-browser and dogfood skills for interactive testing:

/dogfood

Steps:

1. Open the feature in browser
2. Test happy path
3. Test error states
4. Test edge cases (empty states, loading, errors)
5. Capture screenshots for evidence

Phase 5: Final Checklist

Before marking feature complete:

• [ ] Unit tests pass
• [ ] Code review done (subagent or human)
• [ ] No regressions detected
• [ ] UI accessible (if applicable)
• [ ] Documentation updated (if applicable)
• [ ] AGENTS.md updated (if architecture changed)

Workflow Summary

Implement feature
    ↓
Run unit tests → FAIL? Fix first
    ↓
Parallel subagent review (if 3+ files)
    ↓
UI audit + critique (if visual)
    ↓
Browser testing (if interactive)
    ↓
Final checklist → All green? Mark complete

Examples

Example 1: Simple feature (1-2 files)

Input: "Just finished implementing the login form"

Steps:

1. Run: go test ./... (or npm test)
2. Skip subagent review (only 1 file)
3. Run: /audit for accessibility
4. Run: /dogfood to test in browser

Output: "All tests pass. Login form is accessible. Browser testing shows happy path works."

Example 2: Complex feature (5+ files)

Input: "Just finished the payment system — touches 6 files"

Steps:

1. Run: go test ./...
2. Launch subagent review (2 reviewers in parallel)
3. Skip UI audit (backend only)
4. Skip browser testing (no UI)
5. Complete checklist

Output: "Tests pass. Subagent review found 1 issue (missing error handling in payment_handler.go). Fixed."

Example 3: UI feature with accessibility

Input: "Finished the dashboard redesign — new charts and layout"

Steps:

1. Run: npm test
2. Launch subagent review (3+ files touched)
3. Run: /audit → found contrast issue on chart labels
4. Run: /critique → suggested reducing cognitive load
5. Run: /dogfood → all interactive elements work
6. Complete checklist

Output: "Tests pass. Review clean. Accessibility found 1 contrast issue (fixed). Design review suggests simpler chart layout."

Edge Cases

Tests fail

• STOP. Do not proceed to review.
• Fix tests first
• Tell user: "Tests are failing — fix before review"

Subagent review finds critical issue

• STOP. Do not mark complete.
• Fix the issue
• Re-run review on fixed code
• Tell user: "Review found critical issue — fixed and re-reviewed"

UI audit finds WCAG violations

• FIX before marking complete
• Document what was fixed
• If can't fix: document accepted risk with reason

Feature touches <3 files

• Skip subagent review (not worth the overhead)
• Do manual review instead
• Still run tests and checklist

Test Cases

Should activate

• "Test this feature"
• "Run QA on my changes"
• "Check if this is ready to ship"
• "Dogfood the new feature"
• "Review my code before merge"

Should NOT activate

• "Write tests" (writing tests, not running QA)
• "What testing framework do we use?" (question, not action)
• "Fix the flaky test" (fixing, not reviewing)

References

• references/subagent-patterns.md — Subagent task structure patterns