Content
86%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-engineered instruction skill: every phase of the session-sweep workflow is concrete and executable, with explicit guardrails (approval boundary, dedupe-by-intent, no-op rule, immutable inbox writes). The only notable gaps are a slight template/mapping redundancy and the absence of an explicit error-recovery loop after the post-write check.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational instruction with no explanations of concepts Claude already knows, but the structured-record template fields (importance, category, why, context, owner, dependencies, status, confidence) are re-listed in the field-mapping section, and the disambiguation paragraph could be tightened. | 4 / 5 |
Actionability | The write path is copy-paste ready ('npm run issues:add -- --pri P2 --type issue --summary "<title>" --detail "<detail>" --source "session YYYY-MM-DD /ledger sweep"') with an explicit field mapping (P0/P1 → P1, category → Type), an escaping rule, a verification command, and concrete fallback search patterns ('outstanding*', '*issues*', 'TODO.md', 'BACKLOG.md'). | 5 / 5 |
Workflow Clarity | The four-phase sequence (read-only sweep → report with pause for confirmation → filter/dedupe → queue + check) has strong checkpoints, but there is no fix-and-retry instruction for the case where 'npm run check:outstanding-issues' fails, so it falls short of the feedback-loop anchor. | 4 / 5 |
Progressive Disclosure | No bundle files exist; the body is self-contained with clear numbered sections, and its only reference ('Follow every writing rule in the issues skill' → ../issues/SKILL.md) is one level deep and clearly signaled. All inlined content is needed at execution time, so no split is warranted. | 5 / 5 |
Total | 18 / 20 Passed |