Content
85%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 itself is a high-quality orchestrator skill document: lean, dense with executable guidance, and with an unusually well-specified multi-phase workflow including validation and error paths. Its one real defect is bundle integrity — nearly all of the rules and template files the body defers to are missing from the bundle, so the progressive-disclosure promise is unfulfilled.
Suggestions
Ship the referenced bundle files: all ten rules/*.md files and templates/suggestion-pack.md are cited with load timing but absent from the bundle, so every phase's 'Apply <rules/...>' instruction dead-ends. Either include them or inline the minimal content each phase needs.
Inline the worker prompt template (or at least the commit-message format it mandates) in SKILL.md Phase 6 — the 'one commit per applied comment' contract is central, but its concrete shape only exists in the missing rules/handoff.md.
The reference '../../../agents/shared/rules/reviewer-report-ingest.md' escapes the skill directory entirely; move that grammar in-bundle (e.g. under rules/) so the skill is self-contained and the path resolves.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence throughout — no concept is explained (what a PR is, how git works), and every section is operational: a mode-detection table, a gate-score matrix, an eight-phase architecture block, and terse hard rules. There is no padded or over-explained passage left to trim, matching the anchor-5 'every token earns its place' rather than anchor 4's 'minor instances of over-explanation'. | 5 / 5 |
Actionability | Mostly executable guidance: concrete commands ('gh api repos/<owner>/<repo>/pulls/<n>/reviews', 'gw checkout <pr-url-or-number>', 'git status --porcelain', 'gh pr view --json number,url,state,...') and a numeric /confidence decision matrix. It falls short of anchor 5 because the worker prompt template and the commit-message format ('message cites @author + URL') are only pointed at via rules/handoff.md rather than given inline, leaving minor gaps for the most delicate step. | 4 / 5 |
Workflow Clarity | Phases 0–7 are clearly sequenced with explicit parallel/sequential boundaries ('Phases 1–6 run in parallel across PRs; Phase 4 runs sequentially within a PR'), validation checkpoints are present for this batch operation (two-gate validation, 'verify git status --porcelain is empty and HEAD == headRefOid', full check before push), and error-recovery paths are explicit (refuse MERGED/CLOSED PRs, contradicting comments move to surface, auto-rebase is surface-and-stop). This matches anchor 5's 'explicit validation steps; feedback loops; checklists' — the batch-operation cap does not apply since validation is thorough. | 5 / 5 |
Progressive Disclosure | The in-body structure is excellent — a 'Rule | When it loads' table, per-phase read triggers, and one-level-deep references — but scored against the actual bundle, 11 of the 12 referenced files (all ten rules/*.md and templates/suggestion-pack.md) are absent; only references/apply-context.md exists. The navigation design would merit 4–5, but the dangling referenced paths are a major organization gap, not a minor one, placing it at anchor 3 rather than anchor 4. | 3 / 5 |
Total | 17 / 20 Passed |