Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable with a clear, well-validated workflow and real bundle files, but it is over-long and substantially duplicates its own reference materials rather than pointing to them. Trimming the inlined reference content and redundant restatements of the workflow would improve token efficiency and progressive disclosure.
Suggestions
Remove the inlined writing principles, example bug/feature reports, anti-patterns, label options, and review checklist from SKILL.md and point to references/writing_guidelines.md and references/issue_templates.md instead, keeping only the high-level workflow in the body.
Consolidate the three-plus restatements of the workflow (Core Workflow, Important Guidelines, Common Mistakes, Success Criteria, Workflow Summary) into a single canonical sequence plus a brief checklist to cut redundancy.
Keep the scripts reference and submission-preparation templates inline since they are core to execution, but consider moving the full label catalog to the references to reduce inline bulk.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~480-line body re-inlines material already present in the reference files (core writing principles, a full example bug report identical to the one in writing_guidelines.md, anti-patterns, label lists, and the review checklist) and restates the workflow across Core Workflow, Important Guidelines, Common Mistakes, Success Criteria, and a Workflow Summary diagram. It is mostly useful but carries substantial redundant padding. | 2 / 3 |
Actionability | Provides concrete, executable commands with real flags (search_duplicates.sh, fetch_labels.sh --json/--limit), copy-paste-ready heredoc and clipboard templates, exact label names, and a direct GitHub issue URL—fully actionable guidance. | 3 / 3 |
Workflow Clarity | Presents a clearly numbered 6-step sequence with explicit validation checkpoints (confirm-not-duplicate before drafting, ask rather than assume, show draft and get user approval before preparing, review checklist before presenting) and feedback loops for iteration. | 3 / 3 |
Progressive Disclosure | Two real one-level-deep reference files and scripts are well-signaled via a References section, but the body duplicates large amounts of content that already lives in those references (writing principles, example issues, anti-patterns, labels, checklists)—content that should be separate is inline, matching the score-2 anchor. | 2 / 3 |
Total | 10 / 12 Passed |