test-driven-development · ganyuanran — скилл для Claude Code: установка и суть

---
name: test-driven-development
description: "Use when the user explicitly requests strict or test-first TDD, or when the current conversation already contains an explicit `TDD Route: strict` decision from another Aegis workflow."
---

# Execute

→ False-positive entry on a native-direct-skill host? → **Exit immediately unless the user explicitly asked for TDD or the conversation already contains `TDD Route: strict`.**
  In `off` mode, do not start RED / GREEN / REFACTOR from generic bugfix, contract, shared-module, or risky-code wording alone.
  Hand control back to `using-aegis`, `systematic-debugging`, `writing-plans`, or the fast path with verification.
→ Implementing a feature or bugfix under TDD Route `strict`? → **No production code without a failing test first.**
  Gate: medium/high complexity? → route to brainstorming or writing-plans first.
  Mode: default `off` disables automatic TDD, not completion verification; `auto` chooses strict/light/skipped by risk.
  Change Necessity: before strict RED/GREEN enters production edits, confirm the slice really needs a code change.
  Cycle: RED (write test → watch it fail) → GREEN (minimal code → watch it pass) → REFACTOR (clean up → keep green)
  Regression: shared module → related tests. contract change → producer + consumer. core logic → old + new tests.
  Ripple signal hit → cover producer+consumer or real user path before claiming green.
  GREEN proves the currently expressed behavior slice only.
  GREEN does not by itself prove parent-task acceptance, business-value completion, or final completion.
→ Done when: chosen TDD Route is recorded, strict-route tests pass, TDD preflight gate passed when applicable, pre-edit complexity risk was checked for non-trivial source edits, and `verification-before-completion` has fresh evidence.

# Test-Driven Development (TDD)

## Overview

Under `TDD Route: strict`, write the test first. Watch it fail. Write minimal
code to pass.

If you didn't watch the test fail, you don't know if it tests the right thing.

TDD Mode has two values: `off` and `auto`. The default `off` mode disables
automatic TDD routing but never disables `verification-before-completion`.
`auto` lets Aegis choose a `TDD Route` by task risk.

On native-direct-skill hosts, automatic entry must stay anchored to literal
conversation markers such as `TDD Route: strict`, `strict TDD`, `test-first`,
or `RED / GREEN / REFACTOR`, not generic risky-implementation wording.

## When to Use

Only enter this skill after one of these explicit entry signals exists:

- the user explicitly asks for strict TDD, test-first development, or RED / GREEN / REFACTOR
- the current conversation already contains `TDD Route: strict` from another Aegis workflow

Typical strict-route shapes once entry is already justified: new features, bug
fixes, refactoring, behavior or logic changes, interface/data contract changes,
cross-module or shared-module changes, and core logic refactors.

Exceptions (ask your human partner): throwaway prototypes, generated code, config files, pure docs cleanup, read-only diagnosis, comment-only changes.

## TDD Mode and Route

Before source edits, decide:

```text
Aegis Visibility:
- Why this TDD route is strict, light, or skipped:
- What RED/GREEN proves:
- What still needs verification:

TDD Route:
- Mode: auto | off
- Decision: strict | light | skipped
- Strict authority: explicit user/project request | recorded auto decision | not applicable
- Test posture: diagnostic reproduction | post-change regression | strict RED test
- Reason:
- Verification:
```

In `auto`, use `strict` for behavior, bugfix, contract, shared/core, producer /
consumer, persistence, permission, migration, or meaningful regression risk.
Use `light` for tiny low-risk edits with an obvious readback or command check.
Use `skipped` for read-only, docs-only, generated, throwaway, comment-only, or
environment-bound work where TDD does not fit.

In `off`, do not automatically require TDD, create a strict route, or infer one
from risk alone. Explicit user/project TDD requests still apply; risky work may
still need regression coverage and `verification-before-completion` before any
completion claim.
For plan or execution review, `Mode: off / Decision: skipped` is the normal
record unless an explicit user/project strict request overrides it. That record
does not load this skill or turn a diagnostic reproduction into RED. An
approved plan does not supply strict authority by itself.
If this skill was loaded anyway without an explicit TDD request or a visible
`TDD Route: strict` marker, exit instead of improvising an automatic strict
route from risk words alone.

Keep `Aegis Visibility` task-specific: explain the route decision and the
regression boundary, not a generic claim that TDD was used.

## Preflight Gate

TDD is the implementation discipline for an approved behavior or atomic task.
It is not a substitute for task routing, product clarification, or planning.

Before writing tests or production code, stop and route to brainstorming or
writing-plans if the current request has any medium- or high-complexity signal:

- multiple files, modules, pages, screens, services, or owners
- user-visible flows such as navigation, onboarding, checkout, lifecycle, or
  recovery paths
- state transitions, routing rules, API or data contracts, compatibility
  boundaries, migrations, permissions, or persistence
- more than one acceptance path or manual/visual verification requirement
- unclear product behavior, competing constraints, or long-running execution

For these tasks, require a baseline read-set, plan, and atomic tasks before TDD.
High-complexity or ambiguous tasks also need a spec/design review before
planning. Only proceed directly with TDD for low-complexity work whose intent,
owner, compatibility boundary, verification path, and slice goal / success
evidence are already clear.

## Change Necessity

Before strict RED/GREEN enters production code edits, make the code-change
decision visible. Any new source-code path needs this check before RED/GREEN
normalizes it as work to implement. This is the "should code change at all?"
check; it is not a new artifact and does not belong in the `using-aegis` hot
path.

This is behavior-triggered, not prompt-triggered. If strict TDD is about to add
any new source-code path or enter production source edits, expose a natural
readback even when the user did not ask for it. A tiny helper, small guard, new
branch, fallback, adapter, or owner is not exempt. Example: "Code necessity
check: a non-code path is insufficient because <reason>; the minimum change
boundary is <owner/files>, so the decision is code-change."

```text
Change Necessity:
- User-visible need:
- No-change / non-code option:
- Why code change is necessary:
- Minimum change boundary:
- Decision: no-change | docs/config-only | code-change | needs-clarification
```

If the decision is `no-change`, do not write tests or production code for a
non-change. If the decision is `docs/config-only`, route to that narrower
surface and verify it. If the decision is `needs-clarification`, pause before
RED/GREEN. If the decision is `code-change`, carry the minimum boundary into
`TDD Route`, RED, and regression scope.

## Complexity Budget

Before strict TDD on non-trivial work, record the planned complexity budget so
RED/GREEN does not silently normalize a wrong or overloaded owner.

```text
Complexity Budget:
- Artifact class:
- Current pressure:
- Projected post-change pressure:
- Planned governance:
```

Use `using-aegis/references/complexity-governance.md` for shared artifact
classes, pressure signals, and the meaning of planned governance.

## Pre-Edit Complexity Check

Before production code edits, check whether the intended source edit would add
logic to an overloaded or wrong owner. Tiny edits can keep this to one line.

Use `using-aegis/references/complexity-governance.md` for shared pressure
signals and the meaning of `over-budget`.

```text
Pre-Edit Complexity Check:
- Target edit file:
- Existing pressure signal:
- Owner fit:
- Safer edit boundary:
- Decision: edit-in-place | extract helper | add owner file | split task | pause for plan update

Pre-Edit Owner-Fit Decision:
- Edit intent: wiring-only | move-out / extract-first | local-fix-without-new-responsibility | new-responsibility | emergency / compatibility patch
- Owner fit:
- Safer edit boundary:
- Decision: edit-in-place | extract helper | add owner file | split task | pause for plan update
```

If the decision is `pause for plan update`, stop TDD and return to
`writing-plans` or `brainstorming` with the evidence.

If the predicted result is that this slice would push a maintained artifact
over budget and the slice does not also govern that overrun, do not continue
with RED/GREEN as if the task were safely scoped. Pause and update the plan.

When the target edit file is over-budget or mixed-purpose, classify edit intent
before production source edits. `new-responsibility` must not be added in place
by default. `wiring-only`, `move-out / extract-first`, and
`local-fix-without-new-responsibility` may proceed only when they do not add a
new responsibility and the verification boundary is clear. `emergency /
compatibility patch` requires residual risk and a retirement trigger.

When a medium- or high-complexity task needs project records, use configured Aegis workspace support
lazily. Prefer the installed Aegis workspace helper
(`python <aegis-workspace-helper> init --root <target-project-root>`) when it
is available. If the task needs a process trail under `work/`, prefer
`python <aegis-workspace-helper> new-work --root <target-project-root> ...`
so the intent, checkpoint, drift, and evidence paths are indexed and
structurally checkable:

```text
docs/aegis/
  README.md
  INDEX.md
  BASELINE-GOVERNANCE.md
  adr/
  baseline/
  specs/
  plans/
  work/YYYY-MM-DD-<task-slug>/
    10-intent.md
    20-checkpoint.md
    90-evidence.md
    99-reflection.md
```

Do not promote reusable project facts, decisions, specs, or plans into those
directories unless the workflow needs them and no existing project authority
already owns them.

## Red-Green-Refactor

### RED - Write Failing Test

State: input | output | boundary | acceptance criteria. Check existing test coverage first. Write one minimal test showing what should happen.
A minimal test anchors the next behavior slice; it does not by itself define
whole-task completeness unless the parent acceptance is already fully pinned.

<Good>
```typescript
test('retries failed operations 3 times', async () => {
  let attempts = 0;
  const operation = () => {
    attempts++;
    if (attempts < 3) throw new Error('fail');
    return 'success';
  };

  const result = await retryOperation(operation);

  expect(result).toBe('success');
  expect(attempts).toBe(3);
});
```
Clear name, tests real behavior, one thing
</Good>

<Bad>
```typescript
test('retry works', async () => {
  const mock = jest.fn()
    .mockRejectedValueOnce(new Error())
    .mockRejectedValueOnce(new Error())
    .mockResolvedValueOnce('success');
  await retryOperation(mock);
  expect(mock).toHaveBeenCalledTimes(3);
});
```
Vague name; exercises real retry code but asserts only callback count, not the returned result
</Bad>

**Requirements:**
- One behavior
- Clear name
- Real code (no mocks unless unavoidable)
- If a new feature changes user-observable behavior, prefer one minimal
  end-to-end or integration test for the main path before narrower unit tests
- For user-visible work, cover the main journey and the highest-risk experience
  or operational floor before treating unit tests as sufficient
- Add unit tests for core rules, boundary conditions, and error branches

### Verify RED - Watch It Fail

**MANDATORY. Never skip.**

```bash
npm test path/to/test.test.ts
```

Confirm:
- Test fails (not errors)
- Failure message is expected
- Fails because feature missing (not typos)

**Test passes before production edits?** Inspect the reproduction and assertions.
If they miss the requested behavior, correct the test. If the behavior already
exists, reassess Change Necessity; keep useful regression evidence without
claiming a RED/GREEN cycle. Do not break or delete existing code to force RED.

**Test errors?** Fix error, re-run until it fails correctly.

### GREEN - Minimal Code

Write simplest code to pass the test.

<Good>
```typescript
async function retryOperation<T>(fn: () => Promise<T>): Promise<T> {
  for (let i = 0; i < 3; i++) {
    try {
      return await fn();
    } catch (e) {
      if (i === 2) throw e;
    }
  }
  throw new Error('unreachable');
}
```
Just enough to pass
</Good>

<Bad>
```typescript
async function retryOperation<T>(
  fn: () => Promise<T>,
  options?: {
    maxRetries?: number;
    backoff?: 'linear' | 'exponential';
    onRetry?: (attempt: number) => void;
  }
): Promise<T> {
  // YAGNI
}
```
Over-engineered
</Bad>

Don't add features, refactor other code, or "improve" beyond the test.

Fix the real owner of the behavior. Do not add a new fallback, adapter, or
branch unless the debugging or design workflow identifies why it is necessary
and what old path retires.

### Verify GREEN - Watch It Pass

**MANDATORY.**

```bash
npm test path/to/test.test.ts
```

Confirm:
- Test passes
- Other tests still pass
- Output pristine (no errors, warnings)

**Test fails?** Fix code, not test.

**Other tests fail?** Fix now.

### REFACTOR - Clean Up

After green only:
- Remove duplication
- Improve names
- Extract helpers

Keep tests green. Don't add behavior.

### Repeat

Next failing test for next feature.

## Regression Scope

At minimum, run the target test you just changed or added. Broaden regression
based on impact:

- Shared module change -> related module tests
- Interface or data contract change -> producer and consumer tests
- Cross-module behavior change -> integration or end-to-end path
- Core logic refactor -> old behavior regression tests plus new behavior tests
- Ripple Signal Triage fired -> producer+consumer or real user path that proves
  the downstream effect remains bounded

If the current environment cannot run automated tests, state the blocker and provide reproducible manual verification steps.

## Good Tests

| Quality | Good | Bad |
|---------|------|-----|
| **Minimal** | One thing. "and" in name? Split it. | `test('validates email and domain and whitespace')` |
| **Clear** | Name describes behavior | `test('test1')` |
| **Shows intent** | Demonstrates desired API | Obscures what code should do |

## Strict-Route Red Flags - STOP and Diagnose

These red flags apply only after this skill has validly entered under
`TDD Route: strict`. Do not project them onto debugging or regression work
whose route is `light` or `skipped`.

- Code before test
- Test after implementation
- Test passes immediately without an explanation of existing behavior
- Can't explain why test failed
- Tests added "later"
- Rationalizing "just this once"
- "I already manually tested it"
- "Tests after achieve the same purpose"
- "It's about spirit not ritual"
- "Keep as reference" or "adapt existing code"
- "Already spent X hours, deleting is wasteful"
- "TDD is dogmatic, I'm being pragmatic"
- "This is different because..."

Stop and establish which case applies. If you wrote task-owned production code
before the required failing test, remove only that premature implementation
and restart from the behavior test. Preserve pre-existing code and user edits;
if changes are interleaved, isolate ownership before recovery. A test that
passes before any production edit needs the test/Change Necessity diagnosis
above, not code deletion. Missing evidence alone does not authorize a reset.

## Example: Bug Fix

**Bug:** Empty email accepted

**RED**
```typescript
test('rejects empty email', async () => {
  const result = await submitForm({ email: '' });
  expect(result.error).toBe('Email required');
});
```

**Verify RED**
```bash
$ npm test
FAIL: expected 'Email required', got undefined
```

**GREEN**
```typescript
function submitForm(data: FormData) {
  if (!data.email?.trim()) {
    return { error: 'Email required' };
  }
  // ...
}
```

**Verify GREEN**
```bash
$ npm test
PASS
```

**REFACTOR**
Extract validation for multiple fields if needed.

## Strict-Route Verification Checklist

- [ ] Defined input, output, boundaries, compatibility, acceptance criteria
- [ ] Each new behavior has a test observed failing before implementation;
      helpers extracted during GREEN refactoring may use that existing test
- [ ] All tests pass, output pristine
- [ ] Regression: shared/contract/core changes ran related tests
- [ ] Ripple signal hit: downstream or real user path covered
- [ ] If automation blocked → blocker + manual steps documented
- [ ] GREEN treated as local behavior proof only, not final completion
- [ ] If `TaskIntentDraft`, parent plan/spec, or `Slice Card` exists, covered and uncovered scope are explicit before any done claim

Can't check all applicable boxes? Diagnose the gap before claiming strict-route completion.

## Exploration and Emergency Exceptions

Exploratory spikes are allowed only as throwaway learning. When the spike ends,
convert confirmed behavior into tests before formal implementation.

Emergency hotfixes may prioritize the smallest safe repair when delay is more
dangerous than incomplete TDD. Record the reason, keep the change narrow, and
add the missing regression test in the same slice or the next nearest slice.

## When Stuck

Don't know how to test → write wished-for API first. Test too complicated → simplify design. Must mock everything → reduce coupling.

## Debugging Integration

Bug found? Start with `systematic-debugging`: reproduce, trace the owner, and
choose the smallest proof that supports the diagnosis. Under recorded
`TDD Route: strict`, the reproduction becomes the required failing test before
production edits. With `TDD Mode: off` and no strict route, use diagnostic
reproduction and targeted post-change regression as fit the repair; do not
start RED / GREEN by inference.