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.
The body is highly actionable — exact commands, call shapes, paths, and error recovery — with clearly sequenced workflows and strong feedback loops via the error-code table. Its weaknesses are redundancy (port range, skill-fetch pattern, and deploy workflow each repeated multiple times) and the absence of any progressive disclosure structure, with lookup tables and catalogs inlined in an already-long SKILL.md.
Suggestions
State the 17173..17272 port range and the 'never guess the URL, read preview.json' rule once each (e.g. in Hard rules) and reference them from other sections, and consolidate the four repetitions of the searchKnowledgeBase/ui-design fetch pattern into a single canonical block.
Move the error-code recovery table and the state-files table into a references/ file (e.g. references/cli-reference.md) and keep only the codes most likely at runtime (5, 7, 10) inline, turning the ~360-line SKILL.md into an overview with one-level-deep references.
Add an explicit post-rollback verification step to the rollback workflow (e.g. run `cloudbase-sites preview --status` and confirm the version in app.json after `git reset --hard`), since rollback is the skill's one destructive operation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and free of generic concept explanations, but noticeably repetitive: the 17173..17272 port range is stated three times outside the error table, the searchKnowledgeBase/ui-design fetch pattern is repeated in four places, the two-stage deploy is described in its own section and again in hard rule 7, and the browser-preview offer appears in two sections. This fits anchor 3 ('could be tightened') better than 4 ('minor instances that could be trimmed'). | 3 / 5 |
Actionability | Fully executable throughout: exact CLI invocations with flags (e.g. `cloudbase-sites deploy --post --version <n> --access-url <url> [--build-id <BuildId>]`), exact MCP call shapes (`manageApps({ action: "deployApp", serviceName: ..., filePath: cwd, buildPath: "dist", framework: "static" })`), concrete file paths, env-var names, and an error-code table with per-code recovery actions. Copy-paste ready and covers the common cases, matching the 5 anchor. | 5 / 5 |
Workflow Clarity | Clear sequencing with most checkpoints present: the save→deploy→record workflow has explicit gates (nextAction bridge, mandatory BuildId pass-through, pre-flight queryEnv with user confirmation and retry), and the error-code recovery table provides feedback loops for every documented failure. It falls short of 5 because the destructive rollback workflow (`git reset --hard`) has no explicit post-rollback verification step — recovery for code 15 defers to 'check git state' rather than an inline checkpoint. | 4 / 5 |
Progressive Disclosure | Well-sectioned with clear ## headers, but at ~360 lines this is a monolithic SKILL.md with reference-grade material inlined — the 15-row error-code table, the state-files table, and the domain-skill catalog would naturally live in reference files. The one external reference (cloudbase-platform's references/protocols/deployment-share.md) belongs to a different skill's bundle and is conditional, so this skill offers no progressive disclosure of its own; this matches anchor 3 ('content that should be separate is inline') rather than 4. | 3 / 5 |
Total | 15 / 20 Passed |