Content
88%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 excellent, information-dense tool-calling skill that assumes Claude's competence and provides highly actionable field-level guidance, clear multi-step workflows with real validation guards, and well-structured navigation. The main improvement is reducing redundancy between the Caveats and What-NOT-to-do sections.
Suggestions
Merge the 'What NOT to do' section into 'Caveats you must carry into every answer' (or vice versa) — both currently restate open_to_merge-coarseness, pending-unsettled, and no-reviews, costing tokens for reinforcement that one consolidated section could carry.
If the per-tool field semantics grow further, consider splitting the four tool descriptions into a separate reference file and keeping SKILL.md as a tighter overview plus the choosing-a-tool table and high-value chain.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense and information-rich with no padding or explaining of basic concepts (what CI/PRs are), but the 'Caveats you must carry into every answer' and 'What NOT to do' sections restate the same points (open_to_merge coarse, pending unsettled, no reviews) — minor redundancy that could be trimmed. | 4 / 5 |
Actionability | Highly concrete, executable guidance throughout: exact field names ('ci.failing', 'ready_to_merge_seconds', 'author.is_bot'), exact filter logic, explicit aggregation rules ('median and p95, never mean'), and concrete window examples ('date_from=-14d, then date_from=-28d date_to=-14d') — copy-paste ready for an instruction-only tool skill. | 5 / 5 |
Workflow Clarity | The 'high-value chain' gives an explicit 3-step sequence (workflow-health → pull-requests → pr-lifecycle), the 'Choosing a tool' table is a decision checklist, and guards act as validation checkpoints — notably the truncation handling ('shrink date_from until the real set fits under the cap') is a genuine validate→adjust→retry loop. | 5 / 5 |
Progressive Disclosure | Well-organized into clear navigable sections (tools, caveats, choosing a tool, chain, output, what-not-to-do, persisting) with a well-signaled one-level cross-reference to the sibling insights skill, but it is a single inline monolithic document rather than an overview pointing to its own detailed reference files. | 4 / 5 |
Total | 18 / 20 Passed |