Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A lean, opinionated, genuinely actionable TDD reference with concrete gates, a copy-paste question, and precise loop rules — let down mainly by its bundle: both referenced detail files (tests.md, mocking.md) are missing, so the promised examples and mocking rules are inaccessible. The workflow is clear but never assembled into one explicit start-to-finish sequence.
Suggestions
Ship the referenced files or remove the links: neither tests.md nor mocking.md exists in the bundle, so "示例见 [tests.md](tests.md)" and "mocking 规则见 [mocking.md](mocking.md)" are dangling references that break progressive disclosure entirely.
Add one inline example of a good specification-style test (e.g. the "user can checkout with valid cart" case as an actual test snippet) so the '什么是好 test' section is self-contained even if tests.md is unavailable.
Make the loop's validation step explicit — state 'run the test and confirm it fails before writing any implementation' — and consider a short numbered overview (explore → approve seams → red → green → repeat) so the full sequence doesn't have to be assembled from section order.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is tight and opinionated, assumes competence ("TDD 是 red -> green loop" with no padding), and its concrete snippets like "expect(add(a, b)).toBe(a + b)" earn their place. Minor over-explanations could be trimmed — e.g. "因为它不关心 internal structure,所以能承受 refactors" and "你无法测试所有东西" state things Claude can infer — fitting 'efficient; minor instances of over-explanation'. Not anchor 5 because of those few inferable rationales; not anchor 3 because no section is padded or explains fundamentals Claude doesn't know. | 4 / 5 |
Actionability | Highly concrete for an instruction-only skill: a copy-paste-ready question ("What's the public interface, and which seams should we test?"), a specific pre-test gate ("写任何 test 前,先写下要测试的 seams 并与用户确认"), an exact artifact spec ("给每个提议的 seam 加一行说明"), an exact tool invocation ("调用 Skill 工具并指定 codebase-design"), and unambiguous loop mechanics ("先写 failing test,再只写足够让它通过的代码"). Not anchor 5 because the promised examples — "示例见 [tests.md](tests.md)" — point to a file that does not exist in the bundle, leaving the '什么是好 test' section with no example; not anchor 3 because the guidance is far more executable than pseudocode-tier. | 4 / 5 |
Workflow Clarity | A clear sequence is discernible: explore codebase / read GLOSSARY.md / respect ADRs → propose seams and get user confirmation (an explicit human checkpoint) → red before green ("先写 failing test") → minimal implementation → next slice, with refactoring explicitly routed to the code-review skill. Not anchor 5 because the end-to-end order is implied by section order rather than laid out as one explicit sequence, and there is no explicit 'run the test and confirm it fails' step; not anchor 3 because checkpoints (seam confirmation, red-first) are present and explicit rather than implicit. This is not a destructive/batch operation, so no cap applies. | 4 / 5 |
Progressive Disclosure | The bundle contains only SKILL.md — no references/ directory, and neither of the two referenced files ([tests.md](tests.md), [mocking.md](mocking.md)) exists anywhere, so every external reference is dangling and the split-out content is unreachable (GLOSSARY.md is properly guarded with "如果存在"). The in-body structure and signaling are anchor-4 quality, but two-of-two broken references is a navigation failure beyond anchor 4's 'minor organization gaps'; not anchor 2 because the 38-line body is well-sectioned with nothing inlined that belongs in a separate file. | 3 / 5 |
Total | 15 / 20 Passed |