Content
92%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.
This is a well-engineered orchestrator skill: a clearly gated 8-step workflow with an unusually strong safety-scan section, concrete executable search/decision guidance throughout, and textbook progressive disclosure delegating type-specific detail to one-level-deep reference files. The only meaningful weakness is modest token redundancy in the main file, mainly the 'When to Use' section duplicating the frontmatter and output tone/style rules that belong in output-format.md. Overall an excellent example of a research-and-recommendation skill.
Suggestions
Remove the 'When to Use This Skill' section -- it restates the frontmatter description triggers almost verbatim and can be replaced by a single pointer sentence.
Move the tone guidelines, style rules (no em dashes, no emoji), and 'Suggested response' content rules into references/output-format.md, which already owns output formatting, to slim the main file.
Tighten the 'Triage Mindset' section to its bullet list -- the surrounding prose ('AI models have a strong tendency to validate the author's framing') repeats the section's opening point.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and repo-specific throughout -- decision tables, exact label names, and operational rules -- with essentially no explanation of concepts Claude already knows. However, the 'When to Use This Skill' section (~10 lines) duplicates the frontmatter description triggers almost verbatim, the tone guidelines and style rules (em dash/emoji) would fit naturally in the already-referenced output-format.md, and the 'Triage Mindset' section is slightly prose-padded. These are minor trim opportunities rather than noticeable padding, so anchor 4 fits better than 3. | 4 / 5 |
Actionability | For an instruction-only skill the guidance is fully executable: concrete search strings ("regression" "net9.0" "System.Text.Json", '"NullReferenceException" "JsonSerializer.Deserialize"'), named tool invocations (gh issue list --search, github-mcp-server-search_issues), exact label names (needs-author-action, tenet-performance, backlog-cleanup-candidate), precise decision tables (priority, confidence, recommendation criteria), and copy-paste-ready response conventions (fenced code block rule, 'Closing as duplicate of #{number}'). Specific examples cover the common cases for every issue type. | 5 / 5 |
Workflow Clarity | Steps 0-8 are explicitly sequenced with mandatory gates: the safety scan ('MUST complete before any subsequent steps') with a full-stop protocol for prompt injection, duplicate match-quality evaluation before any duplicate claim, a user decision gate ('ask the user to pick one of three outcomes... Do not proceed until they choose'), and confidence calibration with rationale. The skill is plan-only (no destructive or batch operations), and checklists (safety pattern table, anti-patterns, tips) plus feedback rules (verify a merged PR actually addresses the problem; test stale repros against the latest version) satisfy the 5 anchor. | 5 / 5 |
Progressive Disclosure | The body acts as an orchestrator that delegates type-specific detail to eight clearly signaled, one-level-deep references (references/area-label-heuristics.md, bug-triage.md, api-proposal-triage.md, perf-regression-triage.md, question-triage.md, enhancement-triage.md, triage-patterns.md, output-format.md), each introduced with context about what it contains and when to consult it, with the output template externalized. No reference points to another reference, and the inline content (workflow, safety, assessment criteria) belongs in the main file. Note: no references/ directory was present in the reviewed bundle to verify file existence, but the structure as written matches the 5 anchor. | 5 / 5 |
Total | 19 / 20 Passed |