Content
88%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, well-sequenced review methodology with concrete tool examples, an exact output contract, and built-in self-correction loops. Its weaknesses are minor: a few redundant passages that could be tightened, and near-monolithic placement of all policy detail in the single file rather than offloading some to references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with task-specific policy and assumes Claude's competence (no explanations of what PRs or code review are), e.g. 'Breadth first, depth second. The two most common ways to fail are...'. Minor trimming is possible — points like 'A flawless implementation of the wrong thing is still the wrong thing' restate the preceding sentence, and the Tests and Comments bullets run long. Not a 5 because of these few redundant passages. | 4 / 5 |
Actionability | Fully executable guidance: concrete tool calls like 'async () => state.readFile({ path: "<diffPath from your inputs>" })' and 'state.searchFiles({ pattern: ... })', an exact output schema with field-level rules ('path (repo-relative ... not prefixed with /repo/)', 'side (RIGHT for additions)'), and explicit severity/verdict calibration. Specific examples cover the common cases, including the re-review branch. | 5 / 5 |
Workflow Clarity | The method is a clearly sequenced six-step process (frame, judge the approach, enumerate, verify, go deep, prioritize) with explicit validation checkpoints and feedback loops: 'Self-correct: drop candidates that turn out fine; do not report hypotheses you couldn't confirm.' The candidate categories and verdict calibration rules act as checklists. Not a 4 because verification and error-recovery loops are explicit throughout, not merely present. | 5 / 5 |
Progressive Disclosure | The file is well-sectioned (## Documentation changes, ## Your only tool, ## Inputs, ## Method, ## Output) with a clearly signaled one-level reference: 'load the writing-emdash-docs skill before reviewing those files.' No bundle files exist and there is no nested-referencing problem. Not a 5 because at ~100 lines some self-contained policy blocks (e.g. the Changesets subsection) could live in a separate reference file, and the single external-skill reference is the only offloading of detail. | 4 / 5 |
Total | 18 / 20 Passed |