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 unusually actionable and well-sequenced two-mode skill body with strong validation gates, evidence-based review steps, and a Definition-of-Done checklist. Its weaknesses are internal repetition of the log-visibility and feedback rules across multiple sections, and a bundle whose referenced rule and template files are missing, leaving the progressive-disclosure structure hollow.
Suggestions
State the log-visibility and comment-feedback rules once each (in their rule files plus a one-line pointer) instead of restating them in the Non-negotiable section, the Phase 4 gate, Core Principles, the Anti-patterns list, and the 14-line DoD checkbox — this repetition is the main token cost.
Drop the 'Required Reading by Phase' section, which repeats the scaffold phase table's file column verbatim; the earlier table already carries the load-on-demand mapping.
Ship the referenced rule and template files (rules/*.md, templates/*.md) or trim the body to match the actual bundle — 14 of the 15 referenced paths do not exist, so the per-phase 'read this rule file' instructions cannot be followed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows, but the log-visibility rule is restated in four places (Non-negotiable section, Phase 4 gate, Core Principle 9, and the DoD checklist), the comment-feedback rule across three-plus sections, and the 'Required Reading by Phase' table duplicates the scaffold phase table verbatim. This repetition is more than the 4-anchor's 'minor instances' of trimming, though not the pervasive padding of a 2. | 3 / 5 |
Actionability | Guidance is fully executable: copy-paste ready `gh run list ... --jq` metric commands, a concrete bash/YAML snippet for `tee` with `set -o pipefail`, a defined verification grep, and a fully worked review-report format with example output. Specific examples cover the common scaffold and review cases. | 5 / 5 |
Workflow Clarity | The scaffold mode runs five sequenced phases each with an explicit gate ('do not proceed until it passes'), batched Phase-0 intake with repeat-back confirmation, and a Definition-of-Done checklist; review mode mandates per-rule PASS/WARN/FAIL with line-level evidence and n/a-with-reason reporting. Validation steps, feedback loops, and a closing checklist are all present. | 5 / 5 |
Progressive Disclosure | The thin-index design is structurally sound — one-level-deep, well-signaled references organized into a per-phase load-on-demand table. But scored against the actual bundle, only references/decision-tree.md exists; the 14 referenced rules/*.md and templates/*.md paths are absent, so the on-demand loading workflow the body depends on is non-functional. This falls below the 4-anchor's 'references mostly clear' while remaining above the 2-anchor's inlined/buried content. | 3 / 5 |
Total | 16 / 20 Passed |