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.
An exceptionally actionable, well-sequenced review procedure whose every command and pattern is executable and repo-grounded, with strong validation and failure-handling. Its weaknesses are length and repetition — the report shape and read-only rules are enforced redundantly across sections — and a monolithic single-file layout where some reference material could live in bundle files.
Suggestions
State the six-permitted-headings rule and the read-only constraint once each (Step 0 and the "This is a review" paragraph) and refer back to them from later sections, cutting the repeats in "Report shape", the bold-label rule, and Step 5's break-ui passage.
Move the hand-drawn-element lookup table and the per-section grep patterns in Step 4 into a references/ bundle file (e.g. references/element-table.md and references/section-checks.md) linked from the body, slimming the always-loaded SKILL.md to the workflow itself.
Trim rhetorical framing (e.g. "A number nobody measured, written as though someone had, is the one fault this lens must not commit") to plain instructions; the surrounding rules already carry the requirement.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely repo-specific (no explanations of concepts Claude already knows), but at ~630 lines it could be tightened: the six-heading report shape is enforced in three separate places (Step 0, the bold-label rule, "Report shape" plus the closing grep check), "reports and changes nothing" is restated in Step 5 and the break-ui section, and rhetorical passages ("A number nobody measured... is the one fault this lens must not commit") pad beyond instruction. This fits the mostly-efficient-could-be-tightened anchor rather than the minor-trims anchor. | 3 / 5 |
Actionability | Fully executable throughout: a verbatim heredoc report skeleton, exact base-branch resolution commands, per-section grep patterns, a three-step component-lookup procedure with literal git grep invocations, a complete runnable bun -e contrast script including the alpha-flattening helper, and exact finding templates with tier rules — copy-paste ready and covering the common cases. | 5 / 5 |
Workflow Clarity | Steps 0–5 are clearly sequenced with explicit validation checkpoints (the closing grep -c shape check where "Both must print 6"), error-recovery paths ("When a step cannot run", the unmeasured protocol, resume-from-report-file behaviour), and per-step ordering rules (record each finding the moment you have it). The skill is explicitly read-only, so the destructive-operation cap does not apply. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and all external material — DESIGN.md sections, wiki pages, sibling skills' SKILL.md/STANDARDS.md — is referenced one level deep with explicit paths and section names, clearly signalled. It stops short of 5 because the 630-line single file inlines substantial always-referenced material (the hand-drawn-element table, the per-section patterns) that could be split out, leaving the overview role less lean than the clear-overview anchor. | 4 / 5 |
Total | 17 / 20 Passed |