| name | test-authoring |
| description | Use whenever you write new or refactored code during a renovation (frontend or backend) — author tests first (red-green-refactor) and build the automated test coverage the modernized system needs. Cross-cutting; invoke on every code-producing step. Composes with the tdd skill. |
Test Authoring
Cross-cutting skill for every code-producing step of the renovation. New and
refactored code ships with tests written test-first. This is forward-facing
(tests for the new system), complementing behavior-harness (which pins the
old system).
When to use
- Any time
frontend-code-implementation, backend-slice-implementation,
ux-modernization, strangler-backend, or ai-augmentation produces or changes
code.
- This is a gate: no code-producing task is "done" without it.
Composes with existing skills
If a tdd / test-driven-development skill is available in the environment,
invoke it for the red-green-refactor loop — do not reimplement it here. This
skill adds renovation-specific guidance on what to test.
Method
Follow red → green → refactor:
- Red: write a failing test from the acceptance criterion (from
requirements-grilling) before the implementation exists.
- Green: minimal code to pass.
- Refactor: clean up with tests green.
What to test in a web renovation
- New backend units (event handlers, services extracted by the strangler):
unit tests for logic + integration tests across the DB2DB/TCP/API seams.
- New UI flows (single author→approve→result flow, ops dashboard): component
tests for state/interaction + end-to-end tests for the whole flow.
- Realtime paths (live send-status, system-status): tests for the data-link /
event stream, including reconnection and stale-data handling.
- Contract parity: where the new path replaces a legacy one, assert the new
output matches the harness golden master (or a deliberately-updated expectation).
- AI features: deterministic tests around prompt/response plumbing (schema,
guards, fallbacks) — not model output quality; that belongs to
ai-augmentation
evals.
Coverage bar
- Every in-scope acceptance criterion → at least one automated test.
- Regression-prone edges from grilling (bulk, failure/delay, concurrency, partial
send) → explicit tests.
Transition
Once green, hand the change to strict-code-review. The change is not done until
both this gate and the review gate pass.
Output template and helper script
- Korean template: OUTPUT_TEMPLATE.md
- Scaffold script:
scripts/create-test-evidence.sh <repo-root> <target-name>
