`.claude/agents/` was the source of truth, which made every role Claude-Code shaped. Adding a second client meant rewriting each role in that client's syntax and maintaining both copies — the drift this scaffold exists to prevent, one layer up. Roles and skills now live under `.agents/` and render into each registered client. `.claude/agents/`, `.claude/skills/` and `.codex/agents/` are generated; `scripts/sync-agent-integrations.py --check` fails on drift and belongs in CI. The role metadata is portable rather than vendor-named: `reasoning_tier` (deep/balanced/fast/vision), `capabilities`, `mutation`, `invocation`, and an optional `preload_skills`. A client manifest maps those to native syntax and must declare what it cannot express — `codex.yaml` declares `tier_policy: unsupported` and its adapters say so in the file, rather than the tier silently evaporating and leaving the repository to believe it was enforced. The port is behaviour-preserving where it should be and a fix where it should not. Every instruction body is byte-identical — the whole diff to `.claude/agents/` is 18 added lines and zero deletions. What changed is frontmatter that was missing: - four agents (`code-reviewer`, `tdd-guardian`, `dependency-audit`, `pr-creator`) declared no `tools:` and therefore inherited the ENTIRE tool pool, so three review-only agents could edit and write the code they were reviewing. All eight now declare capabilities explicitly. - the six read-only roles gain a non-editing permission mode, so the constraint is enforced by the client rather than by the prompt asking nicely. - `mutation` is now explicit, which records the two roles that genuinely need to write: `pr-creator` (external-write — it pushes a branch and opens a PR) and `dependency-audit` (workspace-write — package managers rewrite lockfiles). `pr-creator` keeps `shell` because opening a PR needs it, but it is now the only agent here with a write mutation and a declared reason for it, instead of one of four with unlimited access by omission.
9.2 KiB
9.2 KiB
name, description
| name | description |
|---|---|
| tdd-workflow | Test-Driven Development workflow with RED-GREEN-REFACTOR cycle, coverage verification, and quality gates. Use when planning or implementing features with TDD. |
TDD Workflow Skill
The TDD Cycle
┌─────────────────────────────────────────────────────┐
│ │
│ ┌─────┐ ┌───────┐ ┌──────────┐ │
│ │ RED │ ──▶ │ GREEN │ ──▶ │ REFACTOR │ ──┐ │
│ └─────┘ └───────┘ └──────────┘ │ │
│ ▲ │ │
│ └──────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────┘
Phase 1: RED (Write Failing Test)
Rules
- Write ONE test that describes desired behavior
- Test must fail for the RIGHT reason
- NO production code exists yet
Checklist
- Test describes behavior, not implementation
- Test name clearly states expected outcome
- Test uses factory functions for data
- Test fails with meaningful error message
- Error indicates missing behavior, not syntax/import errors
Example
// STEP 1: Write the failing test
describe('PaymentProcessor', () => {
it('should reject payments with negative amounts', () => {
const payment = getMockPayment({ amount: -100 });
const result = processPayment(payment);
expect(result.success).toBe(false);
expect(result.error).toBe('Amount must be positive');
});
});
// At this point:
// - processPayment doesn't exist
// - Running test gives: "ReferenceError: processPayment is not defined"
// - This is the RIGHT failure - the function doesn't exist yet
Phase 2: GREEN (Write Minimum Code)
Rules
- Write ONLY enough code to pass the test
- No additional functionality
- No "while you're there" improvements
- Ugly code is acceptable (we'll refactor)
Checklist
- Implementation makes test pass
- No code beyond what test demands
- All existing tests still pass
- Coverage increased for new code
Example
// STEP 2: Write minimum code to pass
export function processPayment(payment: Payment): PaymentResult {
if (payment.amount < 0) {
return { success: false, error: 'Amount must be positive' };
}
return { success: true };
}
// This passes the test - that's enough for now
// Don't add validation for zero, max amounts, etc.
// Those need their own tests first!
Phase 3: REFACTOR (Improve Code Quality)
Rules
- COMMIT before refactoring
- Only refactor if it adds value
- All tests must stay green
- No new functionality
When to Refactor
- Duplicate knowledge (not just similar code)
- Unclear names
- Complex conditionals
- Magic numbers/strings
When NOT to Refactor
- Code is already clean
- Similarity is structural, not semantic
- Would add unnecessary abstraction
Example
// STEP 3: Assess and refactor
// BEFORE REFACTORING: git commit -m "feat: add negative amount validation"
// Assessment:
// - Magic number 0 could be a constant
// - Error message could be a constant
// - But... it's simple enough. Skip refactoring.
// If we had multiple validations:
const VALIDATION_ERRORS = {
NEGATIVE_AMOUNT: 'Amount must be positive',
ZERO_AMOUNT: 'Amount must be greater than zero',
EXCEEDS_LIMIT: 'Amount exceeds maximum limit',
} as const;
const MAX_PAYMENT_AMOUNT = 10000;
export function processPayment(payment: Payment): PaymentResult {
if (payment.amount < 0) {
return { success: false, error: VALIDATION_ERRORS.NEGATIVE_AMOUNT };
}
if (payment.amount === 0) {
return { success: false, error: VALIDATION_ERRORS.ZERO_AMOUNT };
}
if (payment.amount > MAX_PAYMENT_AMOUNT) {
return { success: false, error: VALIDATION_ERRORS.EXCEEDS_LIMIT };
}
return { success: true };
}
Complete TDD Session Example
// === CYCLE 1: Negative amounts ===
// RED: Write failing test
it('should reject negative amounts', () => {
const payment = getMockPayment({ amount: -100 });
const result = processPayment(payment);
expect(result.success).toBe(false);
});
// RUN: ❌ FAIL - processPayment is not defined
// GREEN: Minimal implementation
function processPayment(payment: Payment): PaymentResult {
if (payment.amount < 0) return { success: false };
return { success: true };
}
// RUN: ✅ PASS
// REFACTOR: Assess - simple enough, skip
// COMMIT: "feat: reject negative payment amounts"
// === CYCLE 2: Zero amounts ===
// RED: Write failing test
it('should reject zero amounts', () => {
const payment = getMockPayment({ amount: 0 });
const result = processPayment(payment);
expect(result.success).toBe(false);
});
// RUN: ❌ FAIL - expected false, got true
// GREEN: Add zero check
function processPayment(payment: Payment): PaymentResult {
if (payment.amount <= 0) return { success: false };
return { success: true };
}
// RUN: ✅ PASS
// REFACTOR: Assess - still simple, skip
// COMMIT: "feat: reject zero payment amounts"
// === CYCLE 3: Maximum amount ===
// RED: Write failing test
it('should reject amounts over 10000', () => {
const payment = getMockPayment({ amount: 10001 });
const result = processPayment(payment);
expect(result.success).toBe(false);
});
// RUN: ❌ FAIL - expected false, got true
// GREEN: Add max check
function processPayment(payment: Payment): PaymentResult {
if (payment.amount <= 0) return { success: false };
if (payment.amount > 10000) return { success: false };
return { success: true };
}
// RUN: ✅ PASS
// REFACTOR: Now we have magic number 10000
// COMMIT first: "feat: reject payments over maximum"
// Then refactor:
const MAX_PAYMENT_AMOUNT = 10000;
function processPayment(payment: Payment): PaymentResult {
if (payment.amount <= 0) return { success: false };
if (payment.amount > MAX_PAYMENT_AMOUNT) return { success: false };
return { success: true };
}
// RUN: ✅ PASS
// COMMIT: "refactor: extract max payment constant"
Quality Gates
Before committing, verify:
# 1. All tests pass
npm test # or pytest, cargo test
# 2. Coverage meets threshold
npm test -- --coverage
# Verify: 80%+ overall
# 3. Type checking passes
npm run typecheck # or mypy, cargo check
# 4. Linting passes
npm run lint # or ruff, cargo clippy
# 5. No secrets in code
git diff --staged | grep -i "password\|secret\|api.key"
Coverage Verification
CRITICAL: Never trust claimed coverage - always verify.
# Python
pytest --cov=src --cov-report=term-missing --cov-fail-under=80
# TypeScript
npm test -- --coverage --coverageThreshold='{"global":{"branches":80,"functions":80,"lines":80}}'
# Rust
cargo tarpaulin --fail-under 80
Interpreting Coverage
| Metric | Meaning | Target |
|---|---|---|
| Lines | % of lines executed | 80%+ |
| Branches | % of if/else paths taken | 80%+ |
| Functions | % of functions called | 80%+ |
| Statements | % of statements executed | 80%+ |
Coverage Exceptions (Document!)
# pragma: no cover - Reason: Debug utility only used in development
def debug_print(data):
print(json.dumps(data, indent=2))
/* istanbul ignore next -- @preserve Reason: Error boundary for React */
if (process.env.NODE_ENV === 'development') {
console.error(error);
}
Anti-Patterns
Writing Test After Code
❌ WRONG:
1. Write all production code
2. Write tests to cover it
3. "I have 80% coverage!"
✅ CORRECT:
1. Write failing test
2. Write code to pass
3. Repeat until feature complete
Testing Implementation
// ❌ BAD: Tests implementation detail
it('should call validatePayment', () => {
const spy = jest.spyOn(service, 'validatePayment');
processPayment(payment);
expect(spy).toHaveBeenCalled();
});
// ✅ GOOD: Tests behavior
it('should reject invalid payments', () => {
const payment = getMockPayment({ amount: -1 });
const result = processPayment(payment);
expect(result.success).toBe(false);
});
Writing Too Many Tests at Once
❌ WRONG:
1. Write 10 tests
2. Implement everything
3. All tests pass
✅ CORRECT:
1. Write 1 test
2. Make it pass
3. Refactor if needed
4. Repeat
Skipping Refactor Assessment
❌ WRONG:
1. Test passes
2. Immediately write next test
3. Code becomes messy
✅ CORRECT:
1. Test passes
2. Ask: "Should I refactor?"
3. If yes: commit, then refactor
4. If no: continue
Commit Messages
Follow this pattern:
# After GREEN (feature added)
git commit -m "feat: add payment amount validation"
# After REFACTOR
git commit -m "refactor: extract payment validation constants"
# Test-only changes
git commit -m "test: add edge cases for payment validation"
# Bug fixes (driven by failing test)
git commit -m "fix: handle null payment amount"