Content
86%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.
A lean, well-structured instruction-only skill: efficient token use, a clearly sequenced workflow with security and edit-gating checkpoints, and clean one-level references to the sibling ci-scan skill. The only gaps are minor: no concrete command for the issue-collection step and no guidance for handling cross-check mismatches.
Suggestions
Add a concrete issue-collection command (e.g., a gh issue list invocation with the `[ci-scan]` title filter, repo, and updated-since window) so step 2 is copy-paste executable.
Add a short mismatch-handling branch after step 6 (e.g., "if the report's tally or skipped reasons disagree with the issue set, flag the discrepancy and reconcile before proposing edits") to close the workflow's feedback loop.
State what to do when zero relevant issues are found in the window (e.g., report that and skip to report-only cross-check) so the workflow has a defined path for the empty case.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~40 lean lines: an input table, eight numbered steps, and a bulleted output list, with no padding and no explanation of concepts Claude already knows. Every token earns its place, matching anchor 5. | 5 / 5 |
Actionability | Concrete, executable guidance throughout — "issues whose title starts with `[ci-scan]`", a 30-day default window, enumerated rubric criteria, untrusted-data handling, and an explicit edit gate — but the issue-collection step lacks an executable command (e.g., a `gh issue list` invocation with the filter). Mostly executable with minor gaps, matching anchor 4. | 4 / 5 |
Workflow Clarity | A clear, well-sequenced 8-step workflow with real checkpoints (rubric scoring in step 5, report cross-check in step 6, "Edit files only when the user explicitly asks" in step 8), but no error-recovery loop for mismatches found during the cross-check. This is a read-and-propose workflow rather than a destructive/batch mutation, so the cap-at-3 rule does not apply; it sits at anchor 4 rather than 5, which requires explicit validate-fix-retry feedback loops. | 4 / 5 |
Progressive Disclosure | The skill is under 50 lines, needs no bundle files of its own, and is cleanly organized into Inputs/Workflow/Output sections with one-level-deep, clearly signaled links to the sibling ci-scan skill's files. The simple-skill exception applies and the structure matches anchor 5. | 5 / 5 |
Total | 18 / 20 Passed |