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.
An exceptionally actionable, rigorously sequenced re-review procedure with validation and feedback loops at every risky step. Its weaknesses are verbosity (long single-bullet paragraphs and a Behavior Rules section that duplicates phase rules) and a monolithic structure that inlines templates and sub-procedures a progressive-disclosure design would split into reference files.
Suggestions
Move the self-contained detail blocks — the published-body template (Phase 7), the test-report format spec (Phase 4), the verdict edit-vs-post decision tree, and the follow-up-PR procedure — into reference files (e.g. references/published-body.md, references/test-report.md), keeping SKILL.md a phase-level overview with clearly signaled one-level-deep links.
Remove duplication: the Behavior Rules section restates rules already stated in the phases (e.g. "The push is untrusted code" / "a push does not launder its own contents" appears in both the Security section and Behavior Rules; the approval gates are restated across Phases 6 and 7) — consolidate each rule to one location and cross-reference it.
Break the 200+ word single bullets in Phase 4 (test planning, environment preparation, test-report posting) into numbered steps or short sub-bullets, and tighten the prose — a large fraction of the phrasing could be cut without losing any rule.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Almost all of the body is non-obvious operational procedure Claude would not know (verdict-editing rules, binding cross-checks, bot-wait policy), so it avoids explaining known concepts, but it is noticeably over-written: 250+ word single bullets (Phase 4 test planning, test-report posting), and genuine duplication — "a push does not launder its own contents" appears in both the Security section and Behavior Rules, and the Behavior Rules restate phase rules. It sits between anchor 2's padded sections and anchor 3's mostly-efficient-but-tightenable. | 3 / 5 |
Actionability | Fully executable throughout: exact gh commands with --jq filters (e.g. the prior-review lookup and REST .user.login cross-check), complete markdown templates for the verdict body and test report, exact artifact paths (.artifacts/factory-rereview/pr-<number>-<headSha>.md), and concrete fallback branches for every failure mode — copy-paste ready and covering the common cases. | 5 / 5 |
Workflow Clarity | A clear seven-phase sequence with explicit validation checkpoints and feedback loops: the pre-posting body-file gate, the factory_review_source cross-check with re-run on mismatch, retry-once-after-rejection for the terminal call, the head-must-not-have-moved check before publishing, the adversarial check before every approve, and rebuild-and-rerun on environment failure — matching the top anchor's validate/fix/retry pattern. | 5 / 5 |
Progressive Disclosure | The body is well-sectioned with clear phase headers and its only references (references/categories/README.md and references/archaeology.md in the factory-review skill) are explicit and one level deep, but this skill ships no bundle files and inlines ~61KB of detail — the test-report format spec, published-body template, verdict-editing decision tree, and follow-up-PR procedure are reference-type content living in SKILL.md, matching anchor 3's "content that should be separate is inline". | 3 / 5 |
Total | 16 / 20 Passed |