Content
77%Weight 40%Scale 1-3Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
Highly actionable and clearly sequenced with strong validation feedback loops, but the monolithic inline structure and somewhat verbose, partially repeated shell logic leave room for tighter token economy and file separation.
Suggestions
Extract the repeated reaction-state jq into a single shared script under scripts/ (or one parameterized snippet) to remove duplication between the PR-body and comment reaction blocks.
Move the long GraphQL review-thread query and REST reviews query into reference files, leaving concise invocations in SKILL.md so the overview stays lean.
Tighten the 'Signals' list by merging near-duplicate approval cases (e.g. thumbs-up on PR body vs. on a comment) into a single rule with a sub-bullet.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Assumes Claude's competence (no basic GitHub/gh explanations) and covers non-obvious domain detail, but the ~300-line body with lengthy repeated jq blocks could be tightened; some reaction-state logic appears near-duplicated across surfaces. | 2 / 3 |
Actionability | Provides fully executable gh/jq commands with explicit placeholders and concrete examples — reaction queries, review fetches, and the GraphQL thread query are copy-paste ready. | 3 / 3 |
Workflow Clarity | A clear 9-step loop with explicit validation checkpoints (head-change abort, restart after every push, gate checklist) and feedback loops; the gate-report format and merge condition are unambiguous. | 3 / 3 |
Progressive Disclosure | Well-organized into sections (Hard Rule, Signals, Loop Workflow, Failure Mode) but everything lives inline in one ~300-line file with no bundle files; the long query templates are content that could be split into references. | 2 / 3 |
Total | 10 / 12 Passed |