Content
78%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.
An efficient, opinionated, and highly actionable overview that assumes Claude's competence and encodes clear decision rules (branch selection, ambiguity defaults, capture-on-completion). Its one significant defect is progressive disclosure: the body's two branch links point to LOGIC.md and UI.md, which are missing from the bundle, leaving the core branch instructions inaccessible.
Suggestions
Ship LOGIC.md and UI.md in the skill bundle — the body's central branch decision routes to them, and their absence makes the primary guidance unreachable.
Add one minimal worked example per branch in the body or reference files (e.g. a sample `pnpm` script entry for a UI prototype route, or the skeleton of the tabbed guided-walkthrough HTML) to move guidance from placeholders to copy-paste ready.
Make the completion loop explicit: spell out the validate-with-user step before 'capture' (confirm the verdict, then fold the decision into real code and branch the prototype), so the workflow has a visible checkpoint rather than an implicit one.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and directive throughout — "Prototype 是用来回答一个问题的 throwaway code。问题决定形状" sets framing in one line, and each of the six rules (e.g. "不写 tests,不做超过“能跑起来”所需的 error handling") adds only project-specific policy. No concept Claude already knows is explained; every token earns its place. | 5 / 5 |
Actionability | Concrete, instruction-level guidance dominates: launch via "`pnpm <name>`、`python <path>`、`bun <path>`", a double-clickable single HTML file, "通过 URL search param 和浮动底栏切换", scratch DBs named "PROTOTYPE — wipe me", and commit-to-throwaway-branch capture. As an instruction-only skill this is actionable without code, but it is not 5: the commands are placeholders rather than a worked example (e.g. a sample package.json script or a minimal variant-switching snippet), and branch details are delegated out. | 4 / 5 |
Workflow Clarity | The sequence is clear: identify the question ("来自用户 prompt、周围代码,或在用户在线时直接询问") → pick a branch with an explicit tie-breaking default ("backend module → logic;page 或 component → UI") → build under the six rules → "完成后 capture". Checkpoints exist implicitly (rule 5's expose-state after every action, capturing only "验证过的 decision") but there is no explicit validate-with-user/fix loop, keeping it below 5. | 4 / 5 |
Progressive Disclosure | The structure itself is good — a short overview that cleanly delegates the two branches to well-signaled, one-level-deep links ([LOGIC.md](LOGIC.md), [UI.md](UI.md)). However, neither LOGIC.md nor UI.md exists in the bundle (no references/, scripts/, or assets/ directories are present), so the branch-specific instructions the overview points to are unreachable; navigation is broken rather than easy, which drops it below 4. | 3 / 5 |
Total | 16 / 20 Passed |