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.
A well-engineered review skill: a phased, rule-by-rule workflow with explicit self-validation, concrete commands and repo citations throughout, and a clean split between the in-body rule catalog and the one-level-deep references/rules.md. The only weakness is a handful of redundant justificatory sentences that could be trimmed for token efficiency.
Suggestions
Trim duplicate framing in 'Reviewer discipline' and the summary rules — the severity legend and 'the skill only reports / verdict is advisory' rationale each appear twice; stating each once saves tokens without losing information.
Delete justificatory asides that assume the model needs persuading (e.g. 'Automated reviewers have high false-positive rates') and keep only the directive ('Verify before flagging: read the changed code and confirm the rule applies').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence — 50+ rules rendered as single-line IDs with severity and a citation, no basic-concept explanation — but there are minor instances of over-explanation that could be trimmed (e.g. "Automated reviewers have high false-positive rates," the severity legend explained twice, and "The skill never approves or blocks automatically — the verdict is advisory" restating the earlier "Fix nothing — this skill only reports"). This sits between the level-4 'efficient, minor trimmable over-explanation' and level-5 'every token earns its place' anchors, and the redundant severity/verdict prose keeps it at 4 rather than 5. | 4 / 5 |
Actionability | Guidance is fully executable: exact commands ("git merge-base <base> HEAD", "git grep \"Microsoft.AspNetCore\" -- sdks/dotnet/src"), concrete file citations for every key rule, a per-rule severity taxonomy, a common-pitfalls table, and a copy-paste-ready Markdown output template with worked examples — matching the level-5 'copy-paste ready, specific examples cover the common cases' anchor, clearly above the 'minor gaps' level-4 anchor. | 5 / 5 |
Workflow Clarity | A clearly sequenced Step 0–4 process with explicit validation checkpoints: "Verify the code first" before flagging, Step 3 self-validation ("Dedupe; confirm each finding cites a real rule and a real line ... drop anything not verifiable in the actual diff"), a closing checklist, and clean-diff/reporting rules — the level-5 'explicit validation steps; feedback loops; checklists' anchor. The skill is report-only, so the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | SKILL.md is the overview (rule IDs with one-line summaries, process, output format) and full BAD→GOOD detail and per-rule exceptions live in references/rules.md, which exists (431 lines, with its own TOC), is referenced twice with clear signaling, and is exactly one level deep — matching the level-5 'clear overview with well-signaled one-level-deep references; content appropriately split' anchor rather than the 'minor organization gaps' level-4 anchor. | 5 / 5 |
Total | 19 / 20 Passed |