sme-test
TDD coach: guides users through Given/When/Then test design using a Subject Matter Expert approach. Use when writing new tests, adding test coverage, or learning TDD patterns.
- License 许可
- MIT
- 最近更新
- 2026-09-10
- Stars
- ★ 3
- Forks
- 1
01 前言 Front Matter
- name
- sme-test
- description
- TDD coach: guides users through Given/When/Then test design using a Subject Matter Expert approach. Use when writing new tests, adding test coverage, or learning TDD patterns.
- argument-hint
- [file|function] [--expert|-x] [--auto|-xa]
- user-invocable
- true
02 资源文件 8
- 参考文档adapters/bats/adapter.md4.3 KB
- 参考文档adapters/python/adapter.md3.7 KB
- 参考文档errors/error-handlers.md3.9 KB
- 参考文档prompts/coach-dispatch.md1.7 KB
- 参考文档prompts/gwt-formulation.md2.4 KB
- 参考文档prompts/red-gate.md2.9 KB
- 参考文档prompts/test-writer.md1.7 KB
- 参考文档prompts/three-whys.md2.5 KB
03 SKILL.md 正文
sme-test: Subject Matter Expert TDD Coach
A coaching-driven TDD workflow that teaches you to think in Given/When/Then triples before writing test code.
Arguments: "$ARGUMENTS"
Mode Dispatch
Parse arguments to determine mode:
| Input | Mode | Flow |
|---|---|---|
| (default) | Coach | Full 6-stage coaching flow with 3-Whys dialogue |
--expert or -x |
Expert | Single-turn: paste G/W/T triples directly, skip coaching |
--expert --auto or -xa |
Expert-Auto | Auto-infer triples from code scan with mandatory preview gate |
--auto --yes is not a valid flag combination — the preview gate cannot be
bypassed.
Escape hatches (available at any prompt in expert/expert-auto modes):
- Type
coach→ bail up to full coach mode (3-Whys flow) - Type
paste→ bail sideways to manual triple entry (expert mode)
Coach Mode (Default) — 6-Stage Flow
Stage 1: Dispatch
Read the prompt from prompts/coach-dispatch.md and execute it.
Detect the target code file/function from arguments. Identify the project
language using the adapter DETECT capability. Select the appropriate adapter
from adapters/. Load any existing shared/ memory state from prior sessions.
Subagent: test-coach (Opus) — judgment-heavy, needs deep reasoning for intent extraction.
Stage 2: 3-Whys
Read the prompt from prompts/three-whys.md and execute it.
Ask three progressive "why" questions to surface the real behavioral intent behind the user's test request. Do not accept "test this function" as sufficient — probe until a concrete behavioral contract emerges.
Subagent: test-coach (Opus) — conversational coaching requires judgment.
Stage 3: G/W/T Formulation
Read the prompt from prompts/gwt-formulation.md and execute it.
Guide the user to translate 3-Whys answers into Given/When/Then triples. Each triple must be concrete, testable, and non-redundant. Validate each triple before accepting.
Subagent: test-coach (Opus) — triple validation requires judgment.
Stage 4: Test-Writer Delegation
Read the prompt from prompts/test-writer.md and execute it.
Dispatch a Sonnet subagent (test-writer) with the approved triples and the selected language adapter. The subagent generates executable test code. Stage output for user review before proceeding.
Subagent: test-writer (Sonnet) — structured code generation from approved triples, no deep judgment needed.
Write subagent state to shared/test-writer-output.md. Read all files in
shared/ before dispatching.
Stage 5: RED Gate
Read the prompt from prompts/red-gate.md and execute it.
Dispatch a Sonnet subagent (test-runner) to execute the generated tests. The
tests MUST fail (red state). This gate is mandatory — there is no
override. If tests pass, this is an error (see error class 4 in
errors/error-handlers.md).
Subagent: test-runner (Sonnet) — test execution is deterministic, no reasoning needed.
Write subagent state to shared/red-gate-result.md. Read all files in shared/
before dispatching.
Stage 6: Handoff
Present the failing tests to the user. Offer handoff options:
- Continue manually — user writes implementation code
- Handoff to
/tdd— use the TDD skill for RED→GREEN→REFACTOR - Handoff to
/duper-tdd— TDD in an isolated worktree
Expert Mode (--expert / -x)
Skip Stages 2-3 (no 3-Whys, no coached G/W/T formulation).
- Prompt user to paste Given/When/Then triples in structured format
- Parse triples (no per-triple critique — trust the expert user)
- Proceed directly to Stage 4 (test-writer delegation)
- RED gate (Stage 5) remains mandatory
- Handoff (Stage 6)
Model routing: Sonnet-only — no Opus subagent required. Parsing structured input does not require deep reasoning.
Expert-Auto Mode (--expert --auto / -xa)
Replace Stages 2-3 with automated triple inference.
-
Code scan: test-coach (Opus) scans target file, infers triples from function signatures and docstrings
-
Provenance tags: Each inferred triple carries a tag identifying which function/signature it was inferred from
-
Cap: Maximum 8 triples per invocation (hard cap)
-
Mandatory preview gate: User sees all inferred triples and must explicitly:
- (k)eep all — accept all triples
- (d)rop N — remove specific triples
- (a)dd — add custom triples
- (abort) — cancel entirely
-
Proceed to Stage 4 → Stage 5 (RED gate mandatory) → Stage 6 (handoff)
Model routing: test-coach on Opus for inference quality (code understanding + triple generation).
Subagent Coordination
All subagents coordinate via the shared/ memory directory pattern (per
CLAUDE.md):
| Subagent | Model | Stages | Writes to shared/ |
|---|---|---|---|
| test-coach | Opus (coach), Sonnet (expert) | 1-3 | shared/coach-state.md |
| test-writer | Sonnet | 4 | shared/test-writer-output.md |
| test-runner | Sonnet | 5 | shared/red-gate-result.md |
Rules:
- Each subagent writes only its own file in
shared/ - Each subagent reads all files in
shared/on startup - Main skill agent manages cleanup and compacts to
COMBINED.mdafter subagent completion - No subagent modifies
COMBINED.mddirectly
Error Handling
See errors/error-handlers.md for the 5 error classes and recovery paths:
- Input errors — malformed triples, unsupported language, missing target
- Environment errors — test runner not installed, wrong version, missing deps
- Subagent errors — timeout, unexpected output, coordination failure
- RED-gate errors — tests pass when they should fail (false green)
- Safety-refusal errors — subagent refuses to generate test
Adapter Contract
All language adapters implement 4 capabilities: DETECT, GENERATE, RUN, EXPLAIN.
v1 adapters:
adapters/python/adapter.md— pytestadapters/bats/adapter.md— bats (N-block parametrize limitation documented)
04 同仓库更多 Skills
visual-aid
Use when the user asks for a quick visual aid, visual explainer, concept card, one-pager, poster, cheat sheet, infographic, visual summary, summary card, or mini explainer page — especially from input like a block of text, attached images, an attached song (lyrics/audio), a web link, a CSV/dataset, a math formula, a code snippet to walk through, a tutorial's steps, a timeline, or an A-vs-B comparison. Triggers include "make me a quick visual aid", "turn this into a one-pager / cheat sheet / explainer", "summarize this visually", "give me a visual of", "explain this on a single page", "make a concept card for", "whip up a page about", "build a mini page that shows", "something I can paste into a doc explaining X", or any request to render a topic/text/link/image-set/song/data into a standalone page. Prefer this skill over frontend-design when the ask is an explanatory single-file page (not a production UI), over playground when the output is static information display (not an interactive configurator), and over a Markdown/code answer when the user explicitly wants a *page* (HTML). Produces a single self-contained HTML file that opens on double-click without a server.
tesseract
Use when you've been poking at the same file, concept, or problem across many sessions and want to step outside time to see its whole arc — or when you want to leave a note for next-session-you. Triggers: 'I've been going in circles on X', 'what have I learned about X', 'what did past-me say about X', 'I keep rediscovering this'.
tdd
Invoke test-driven development as an explicit one-at-a-time micro-cycle — one test, one piece of functionality, refactor, then repeat for the next behavior
tdd-cat
TDD workflow with subagent-driven-development — /tdd composed with /cat, plus the A/B TDD execution-mode axis (split-phase vs micro-cycle) asked as a third question in /cat's combined pre-dispatch question
task-list
TaskCreate / TaskUpdate / render the current TaskList in response to `/task-list` (with or without flags). The skill body executes immediately on the syntax — no permission-seeking, no echoing the list back as plain text.
sync
Sync files between two local directories, repos, or git worktrees. Five modes (plan/interactive/push/pull/mirror), .gitignore-aware, Claude-workflow allowlist via --claude, dropbox-restore-style per-directory prompts.