Content
77%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.
The body is an unusually actionable, well-sequenced workflow with exact commands, complete examples, explicit validation, and a hard approval gate — the operational core is excellent. Its weaknesses are length and layout: the giant config-mapping table and confidentiality detail are inlined where one-level-deep reference files would keep the overview lean.
Suggestions
Move the long-tail rows of the configuration mapping table (CI comment filtering, default-branch review, draft-PR handling, budget routing) into a references/ file such as CONFIG-MAPPING.md, keeping the dozen most common mappings inline.
Extract the detailed Confidential material routing checklist into its own reference file, leaving a short in-line summary plus the "ask whether that is acceptable" instruction.
Deduplicate the binary_evidence and draft-PR guidance that currently appears in both the mapping table and the "Rules that matter" list.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense and never explains concepts Claude already knows (ruling out 1-2), but at ~550 lines it could be tightened: several table rows are dense multi-line paragraph dumps (e.g. the post_commit_quiet_period and comment_min_confidence rows), and content repeats between the mapping table and the "Rules that matter" list (binary_evidence and draft-PR handling each appear twice). Not 4 because the redundancy and length are more than minor over-explanation. | 3 / 5 |
Actionability | Guidance is fully executable: exact commands ("roborev doctor --from-skill --json", "git check-ignore --no-index -v review-rubrics/facts.md"), a complete .roborev.toml example, a complete rubric template, and concrete gotchas like "Write a literal `{{` as `{{"{{"}}`". The legal-brief example covers the common non-code case. | 5 / 5 |
Workflow Clarity | A clear end-to-end sequence (Look first → conversation → playback → configuration → proposal → build and verify → trial review → hand off) with explicit validation checkpoints (config validate, doctor, git check-ignore, git rev-parse --verify, git show for sources), a hard approval gate, and a tune-rerun feedback loop ("Show each rubric change, get approval, commit it, and request a new review"). | 5 / 5 |
Progressive Disclosure | No bundle files exist, so everything is inlined in one ~550-line SKILL.md. Section headers are clear (better than anchor 2's minimal structure), but large blocks that clearly belong in reference files are inline: the 30+ row configuration table and the detailed confidentiality routing rules. Not 4 because the split-out opportunity is substantial, not a minor organization gap. | 3 / 5 |
Total | 16 / 20 Passed |