Content
92%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 exemplary skill body: dense, non-obvious Kopai-specific guidance with explicit per-step completion checkpoints, error-recovery loops at every failure mode, and clean one-level-deep reference structure (recipes.md, artifact-capabilities/mcp.d.ts). The only gap keeping actionability from the top anchor is the worked-example JS fragment, which depends on page-side identifiers that are specified in prose but not executable as written.
Suggestions
Make the worked-example JS runnable as written, or mark the placeholders explicitly (e.g. a comment noting RANK/CHART are the JSON queries above and watchSection is defined by the described wrapper contract), so the example is copy-paste adaptable without inference.
Consider adding one minimal complete `watchSection` implementation in the recipes reference, since the doc already fully specifies its behavior (reads result.payload.data, updates 'last updated' from result.cache?.storedAt, routes error events, returns unsubscribe).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean throughout: it assumes competence, explains only non-obvious Kopai specifics ('Ranking by error count keeps a service with one failed span from topping the chart', 'Polling faster than the buckets change returns the same rows'), and delegates the runtime API contract to the artifact-capabilities skill rather than restating it. No padding or explanations of concepts Claude already knows. | 5 / 5 |
Actionability | Guidance is highly concrete: exact connector/tool names, two complete copy-paste-ready KopaiQuery JSON payloads, exact error-state messages with URLs, and a `refetchInterval` formula ('5m → 300000'). It falls short of 5 only because the JS example is a fragment relying on undefined page-side identifiers (`RANK`, `CHART`, `watchSection`, `renderChart`) rather than fully executable code, though `watchSection`'s behavior is precisely specified in prose. | 4 / 5 |
Workflow Clarity | Six sequenced steps each end in an explicit 'Done when' checkpoint, with feedback loops throughout: repair from `invalid_input` issues and re-call, zero-rows sends you back to step 3 with 'never publish an empty chart', near-match connectors get a rename instruction, and publishing is gated on step 1's exact-name check. This matches the top anchor including validation steps and error-recovery loops. | 5 / 5 |
Progressive Disclosure | The SKILL.md is a well-structured overview: workflow plus one worked example inline, query recipes split into a clearly linked one-level-deep references/recipes.md (verified to exist and be well-organized), and the runtime API contract delegated to the artifact-capabilities skill's `mcp.d.ts` with explicit instruction to load it before step 5. Navigation is easy and nothing that belongs in a separate file is inlined. | 5 / 5 |
Total | 19 / 20 Passed |