Content
56%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.
The body is highly actionable and unusually disciplined about verification — an engineering constitution with explicit static and browser-level checkpoints, deploy-then-verify MCP calls, and a no-papering-over rule. Its weaknesses are material redundancy (two sections largely duplicate two others, and key warnings are repeated verbatim) and a broken disclosure network in which the two referenced bundle files, frameworks.md and browser-testing.md, are missing.
Suggestions
Merge 'When to use this skill' into the Activation Contract and 'Core workflow' into 'How to use this skill' — they duplicate the same use/do-not-use lists and validation mandates — and state the History-mode 404 gotcha and the ui-design load-first warning once instead of twice.
Ship the referenced bundle files frameworks.md and browser-testing.md (or inline their essential workflows), since the body directs the agent to them for framework conventions and the concrete browser validation loop, and neither exists in the bundle.
Consolidate the triple-stated agent-browser validation requirement into the Engineering constitution and reference it from the workflow sections with one line each.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly a quarter of the body is redundant restatement: 'When to use this skill' duplicates the Activation Contract's use/do-not-use lists, 'Core workflow' repeats 'How to use this skill' (browser validation and CloudBase-as-branch appear in both), the agent-browser mandate is stated three times, and the History-mode 404/NoSuchKey explanation is given in full twice. This matches anchor 2's 'several unnecessary explanations or padded sections' rather than 3's occasional tightening, though it avoids anchor 1 because there are no basic-concept explanations and every line is directive. | 2 / 5 |
Actionability | Concrete and executable throughout: exact commands ('tsc --noEmit', 'npm install @cloudbase/js-sdk'), copy-paste MCP calls (manageHosting({ action: "setWebsiteDocument", indexDocument: "index.html", errorDocument: "index.html" }) with a queryHosting verify step), a runnable SDK init snippet, and specific file patterns to inspect ('src/lib/backend.*', 'src/lib/auth.*'). It falls short of 5 because the central browser-validation workflow is deferred to 'browser-testing.md', which is absent from the bundle, leaving the most important concrete workflow unavailable. | 4 / 5 |
Workflow Clarity | Validation checkpoints are explicit and layered — static checks (tsc/eslint/build/tests) plus a browser layer with 'Only after both layers pass may you say the task is done', a rule to disclose exactly which step is unverified, and a deploy-then-verify loop, so the destructive-operation cap does not apply. Not 5 because the body is organized by constitution/topic rather than one sequenced end-to-end process, and the detailed browser validation loop lives in the missing 'browser-testing.md'. | 4 / 5 |
Progressive Disclosure | In-body structure is good with plainly signposted one-level-deep pointers ('General React / Vue / Vite guidance -> frameworks.md', 'Browser flow checks or page validation -> browser-testing.md'), but neither referenced file exists in the bundle (no references/ directory at all), so the two primary go-deeper paths are dead links; the ~50-line CloudBase operations section is also inlined. This sits at anchor 3 rather than 4 because navigation breaks exactly where the skill directs the agent to read more, and above 2 because the SKILL.md itself is well-sectioned with prominent, non-buried references. | 3 / 5 |
Total | 13 / 20 Passed |