Content
77%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 exceptionally actionable, well-sequenced procedure whose commands, JSON templates and validation gates leave nothing ambiguous about what to run, when to stop, and what to record. Its costs are density and self-containment: the coverage and ledger rules are stated redundantly inline, and material that belongs in one-level-deep reference files (schemas, receipt rules) is embedded in the body.
Suggestions
Move the proposal/run JSON schemas and the coverage-receipt classification rules into a reference file (e.g. references/ledger-schema.md) and keep a compact summary plus a clearly signaled link in SKILL.md.
Tighten the Inspect coverage section by stating each `through`/receipt rule once — the conditions currently appear in both the prose paragraph and the `record` refusal list.
Replace the long prose list of `record` refusals with a short table (refusal → reason → action), which would cut tokens without losing any rule.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (no concept explanations, no padding) but restates the coverage/`through` rules repeatedly — the receipts paragraph ('It accepts a partial `coverage.through` only when every source not read in full is a `complete` or `capped` source whose content you read and names its own `through` at or after it...') and the record-refusal list re-cover the same ground. Mostly efficient with sections that could be tightened, matching anchor 3 rather than the minor-trim level of 4. | 3 / 5 |
Actionability | Guidance is fully executable: copy-paste-ready commands with real flags (`node scripts/lenses/ledger.mjs due <LEARNER>`, `gh pr list --state all --search "updated:>=<time> sort:updated-asc" --limit 100`), a complete JSON run-record template, and a precise field-by-field proposal schema with exact allowed values. Matches anchor 5; for an instruction-only skill the absence of code is not penalized because the commands and schemas are exact. | 5 / 5 |
Workflow Clarity | The sequence (Guards → Inspect → Record → Run output) is explicit with hard validation checkpoints at every stage: `"due": true`/`"allowed": true` gates before any work, `record` refusing malformed, late, out-of-window or oversized runs, per-source coverage receipts, and explicit error recovery (push refused → report and stop, slot stays due; revision semantics for changed repeats). This batch-write skill has the feedback loops anchor 5 requires, well above anchor 4's 'minor validation gaps'. | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent; `scripts/lenses/ledger.mjs` is a repository path, not a bundle file), so the 200+ line body inlines material — coverage-receipt rules, the proposal schema, the ledger refusal rules — that would fit one-level-deep reference files. Headers are clear and well-ordered, but there are no references to signal and the inline bulk is substantial, matching anchor 3 rather than the mostly-appropriate placement of 4. | 3 / 5 |
Total | 16 / 20 Passed |