code

/code - TDD Implementation

Implement via the test-first RGRC cycle, leaving production code that satisfies each Spec Test Scenario (T-NNN) and passes the quality gates. When no Spec exists, still leave each change as test-verified code.

Rationalization Counters

Rationalization	Counter and action
This is too simple for TDD	Simple changes hide regressions most. Write one test line first
I'll add tests later	Deferral compounds into debt. Write the test now, then implement
This is just a refactor	Untested refactors are the top cause of silent regressions. Confirm existing tests are green before starting
Existing tests already cover	Don't assume. Run the Red phase to confirm coverage
Testing this would be too slow	A slow test still beats a production bug. Write it anyway
Red test doesn't need verification	An unrun Red is blind. Run it; confirm the failure matches the intent

Input

$ARGUMENTS contains the implementation description. Required; prompt the user if empty.

SOW Context

See ${CLAUDE_SKILL_DIR}/../_lib/sow-resolution.md

Scope Guard

After reading OUTCOME.md and SOW, check (a) Phase file counts and (b) outcome alignment. If any Phase has Files ≥ 5, stop and ask user to split via /think before proceeding. If the implementation steps into Non-goals or conflicts with Constraints from OUTCOME.md, stop and confirm with user. If no SOW exists and $ARGUMENTS implies ≥ 5 files, suggest running /think first.

Execution

When the RGRC cycle writes framework or library API code, follow the pinned version's official docs rather than memory. Apply ~/.claude/rules/development/SOURCING.md to prevent version drift.

Step	Action	Detail
0	Outcome Anchor	Read .claude/OUTCOME.md; if absent, generate the stub via /outcome
1	SOW Context	Detect and read SOW/spec → Scope Guard
2	Spawn generator-test	subagent_type: generator-test, run_in_background: true
3	Receive test results	TaskOutput (wait for completion before implementation)
4	Test Review	Spawn reviewer-coverage; resolve wrong assertions / weak or tautological / intent mismatches before implementing (advisory)
5	RGRC cycle	gates auto-retry per Green. Meet the target (0 failures) paired with the preserved invariant (no test deleted/skipped)
6	Review Gate	Spawn reviewer-readability
7	Storybook Phase	Conditional. See ${CLAUDE_SKILL_DIR}/references/storybook-phase.md
8	E2E Phase	Conditional. See ${CLAUDE_SKILL_DIR}/references/e2e-phase.md
9	Quality Gates	See use-workflow-code. If a SOW exists, manually confirm AC are met

Spec Evolution

T-NNN is a T-\d{3} three-digit zero-padded spec scenario ID (e.g. T-001), embedded in the test function name, describe/it string, or inline comment.

When a new requirement surfaces during implementation (edge case, error handling, integration concern), update the Spec first, in this order.

1. Add T-NNN to the Test Scenarios table in spec.md
2. Write the test referencing that T-NNN in its name/comment

Every test maps to a T-NNN; never add a test without a Spec trace. evaluator-test uses these mappings to compute coverage and other quality metrics.

Error Handling

Error	Action
generator-test timeout	Leader generates tests directly
generator-test produces 0 tests	Verify spec exists, ask user
reviewer-coverage cannot launch	Skip; Leader inspects the tests directly (advisory)
reviewer-readability cannot launch	Skip; Leader reviews directly (advisory)
Test runner not detected (non-JS/TS)	Re-spawn with the test runner specified in the task prompt
Quality gates fail	Fix issues before commit
Evaluator metric below threshold	Fix uncovered/excess/duplicate/granularity/intent issues
Evaluator timeout	Skip gate, log warning
Spec not found	Proceed without T-NNN trace, skip Test Quality gate (or ask user to create spec)
agent-browser crash	Skip E2E, advisory, continue
Dev server unreachable	Skip E2E, advisory, continue
E2E tests fail	Advisory (do not block)
Storybook phase error	Skip phase, advisory, continue