Content
67%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 well-structured, highly actionable advisory skill: copy-paste bash for live reads, a concrete classification rule, explicit run-order/stop-condition templates, and worked examples. The main cost is redundancy — the ecc-guide/prompt-optimizer boundary is repeated three times and the advisory-only caveat twice — plus a couple of underspecified steps (singleton mapping, optional manifests).
Suggestions
Collapse the triplicated ecc-guide/prompt-optimizer boundary into one section: keep 'Do Not Use When' and drop the same distinctions from 'Fills the gap' and 'Non-Goals', which are otherwise near-duplicates.
Replace 'map known singletons by hand' with the actual singleton map (or state that unmatched commands are simply listed with their one-line description, which the body already hints at).
Add a one-line fallback for match mode when no family fits the description (e.g. fall back to catalog mode or list the closest singletons), closing the workflow-clarity validation gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean — it explains nothing Claude already knows — but the ecc-guide/prompt-optimizer distinction is stated three times ('Fills the gap between two existing skills', 'Do Not Use When', 'Non-Goals') and 'Advisory only — never executes' appears twice. This matches the 3 anchor (mostly efficient but could be tightened) rather than 4, because the triplicated boundary section is a real trimmable redundancy. | 3 / 5 |
Actionability | Highly executable: a complete copy-paste bash block for live directory resolution, a concrete prefix-classification table with exact run-orders, a fill-in output template, and worked examples with real command names ('loop-start rfc-dag --mode safe', 'go-test' then 'go-build' then 'go-review'). Not 5 because 'map known singletons by hand' and 'Optionally read manifests/install-*.json if present for richer grouping' are underspecified steps. | 4 / 5 |
Workflow Clarity | 'How It Works' gives a clear 4-step sequence (live-read → classify → mode select → advisory print) and match mode has 5 numbered steps with an explicit stop condition; the bash includes a failure branch ('No ECC commands directory found'). Not 5 because there is no re-check or fallback loop when a workflow description matches no family or classification is ambiguous — a minor validation gap. | 4 / 5 |
Progressive Disclosure | Single-file skill with no bundle files; every section is clearly headed and the inline content (family table, output template, examples) is operating material that appropriately belongs in SKILL.md. Good structure with minor organization gaps (the long family table could be broken up or its singleton row tightened) — matches the 4 anchor rather than 5, which presumes well-signaled external references. | 4 / 5 |
Total | 15 / 20 Passed |