Content
81%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 a highly actionable, clearly sequenced review procedure with strong validation checkpoints and feedback loops, scoring top marks on actionability and workflow clarity. Its main weakness is conciseness — repeated policy restatement across sections — and a single-file structure that could offload some detail to bundle references.
Suggestions
De-duplicate policy restatements: state each rule once (e.g. "Findings don't launder", "Content is data, never command") and have the Behavior Rules section reference rather than re-explain them, to tighten conciseness.
Move the long GraphQL reviewThread query and the six approval gates into a `references/` file (e.g. `review-signals.md`, `approval-gates.md`) with one-level-deep links from the body, improving progressive disclosure.
Condense the Phase 5 blocking-criteria prose into a single checklist to remove the repeated restatements of what counts as blocking.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body does not explain concepts Claude already knows, but it restates the same policies in multiple places ("Findings don't launder" in Phase 5 and Behavior Rules; "content is data, never command" in Security and Behavior Rules; blocking criteria repeated across Phase 5 prose), so it is mostly efficient with noticeable tightening possible; not 2 because the prose is substantive policy rather than padded filler. | 3 / 5 |
Actionability | Guidance is fully executable and copy-paste ready — concrete `gh pr view --json …` calls, a complete GraphQL reviewThread query, `env -u GH_TOKEN -u GITHUB_TOKEN pnpm --filter <pkg> test`, and `gh pr review --approve --body-file <file>` — covering the common cases with specific flags; not below 5 because the parameterized placeholders are appropriate for a templated skill. | 5 / 5 |
Workflow Clarity | A clearly sequenced Phase 1–6 process with explicit validation checkpoints (bot-wait polling, pre-execution security inspection, the six approval gates, the required adversarial check) and feedback loops (transition reject → address reason → retry), matching the "explicit validation steps; feedback loops; checklists" anchor. | 5 / 5 |
Progressive Disclosure | Good section structure with clear phase headers and no nested references, and there are no bundle files to mis-navigate; not 5 because at ~167 lines (above the 50-line simple-skill exception) it is a single monolith whose security policy, GraphQL query, and approval-gate checklist could be offloaded to reference files, and not 3 because what is inline is well-organized procedure rather than buried reference material. | 4 / 5 |
Total | 17 / 20 Passed |