Content
81%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, policy-dense instruction skill: workflows are explicitly sequenced with verification feedback loops and safety gates, and lookup routing is clean. The main costs are heavy policy prose that taxes the context window and near-total delegation of concrete contracts to external documentation, leaving minor actionability gaps.
Suggestions
Rewrite the "Documentation fallback" and "Before designing or editing" sections as short numbered imperatives; several layered sentences mix multiple rules per line and could be compressed to roughly half their tokens without losing meaning.
Add one minimal inline example per workflow (e.g., a tiny `ext.yml` manifest snippet with the required fields, or a sample `clacky ext verify` error with its `code`/`file`/`hint` and the fix) so the most common cases are actionable without fetching external docs.
Move the detailed Engineering-checks rules into a clearly-signaled reference section or file, keeping a short checklist inline in SKILL.md, so the overview body stays lean while the full criteria remain discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and policy-driven with no tutorial padding about concepts Claude already knows, so it clears the verbose anchors. Not 5 because sections like "Documentation fallback" and "Engineering checks" carry layered, hard-to-parse policy prose (e.g., "Reuse verified text still in context. For a named interface, obey the system prompt's strict lookup budget") that could be stated more directly with fewer tokens. | 4 / 5 |
Actionability | Concrete, executable anchors are present throughout: `clacky ext new <id>` with `--full`, `clacky ext verify` reading its `code`/`file`/`hint`, `clacky ext list`, `~/.clacky/ext/local/<id>/`, and named UI primitives (`form-input`, `Clacky.Modal`, `var(--color-*)`). Not 5 because the skill deliberately delegates all signatures, examples, and reload specifics to external documentation, leaving a few common cases without an inline worked example. | 4 / 5 |
Workflow Clarity | Scaffold, Debug & verify, Publish, and Handoff are clearly sequenced with explicit validation checkpoints and feedback loops: run tests → `clacky ext verify` → resolve errors via actual `code`/`file`/`hint` → verify the visible feature, plus a re-applied Engineering-checks checklist. Destructive/paid/publish operations are gated on explicit approval, satisfying the destructive-operation cap. Not 4 because validation is explicit at every stage rather than implicit. | 5 / 5 |
Progressive Disclosure | The "Capability → reference" table is a well-signaled, one-level-deep routing index ("Read the pages needed for this task, not every reference indiscriminately"), and no bundle files exist to nest. Not 5 because a substantial amount of dense policy material (documentation fallback rules, engineering checks) lives inline in SKILL.md itself rather than being split into navigable sections or files, making the overview heavier than ideal. | 4 / 5 |
Total | 17 / 20 Passed |