Content
87%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a well-structured, lean router with concrete commands and explicit file/parameter dispatch, and it practices strong progressive disclosure by keeping all detail in one-level-deep reference files. Its one weakness is workflow clarity: it orchestrates destructive auto-fix modes without surfacing any explicit verify-before-submit checkpoint in the body itself.
Suggestions
Add an explicit validation checkpoint for destructive auto-fix paths in the body — e.g., a step requiring fixes to be re-verified (tests pass / verifier confirms) before results are submitted — so workflow_clarity is not capped at 2 for batch/destructive operations.
Reduce the minor redundancy between the routing table and the priority-rule blockquote, or fold the precedence rule directly into the table's Condition column, to tighten conciseness further.
Briefly state in the hand-off that target reference files are responsible for their own validation/feedback loops, so readers understand where verification lives rather than inferring it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — it never explains what a PR, git, or a worktree is, and uses terse routing tables and direct instructions ("Each `→` means: `Read` the target file and follow it as the sole remaining instruction"); the only mild redundancy is the priority-rule blockquote, which disambiguates a genuinely ambiguous state-vs-intent precedence and earns its place. | 3 / 3 |
Actionability | Guidance is concrete and executable: exact pre-check commands (`git branch --show-current`, `git status --porcelain`), a routing table mapping conditions to specific reference files, and explicit parameter passing ("Pass `$ARGUMENTS` to the target file... also pass `FIX_MODE` (low / low_medium / full)") — copy-paste ready rather than abstract direction. | 3 / 3 |
Workflow Clarity | The routing sequence is clearly ordered top-to-bottom (pre-checks → first applicable rule → question → hand-off with FIX_MODE), but the skill routes destructive/batch auto-fix operations ("auto-fix all", "auto-fix low & medium risk") without any explicit validation or verify-before-submit checkpoint in the body — verification is only delegated to the reference files, which caps this at 2 per the destructive-operation guideline. | 2 / 3 |
Progressive Disclosure | The body is a clean overview/router that points to one-level-deep, clearly signaled references (cherry-review-guidance.md, diagnosis.md, pr-review.md, local-review.md, teams-review.md), all of which exist as real bundle files; detailed review logic is appropriately split out of the body with easy table-based navigation and no nested reference chains. | 3 / 3 |
Total | 11 / 12 Passed |