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.
Excellent procedural content: every check is executable with exact commands, regexes, and verbatim messages, and the workflow is fully gated with explicit fail/skip semantics before a destructive submit. The only real costs are mild repetition of the SKIPPED/air-hatch rationale and a stub-template section that could be split into a reference file.
Suggestions
State the SKIPPED/air-hatch rationale once (e.g. in Step 0 or the Stub templates section) and reference it from Steps 2, 3, 6, and 7 instead of repeating it, saving a few dozen tokens.
Move the two stub templates into a `references/stub-templates.md` file and keep only the paths and one-line purpose inline in Step 0, so the ~40-line template block is loaded only when the user picks Quick submit.
Step 4's block/warn blockquotes embed literal \n sequences ("...will not see them.\n\nPlease run:"); use real markdown line breaks or a fenced block so the message reads correctly when printed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with commands, exact regexes, and verbatim prompts with almost no explanation of concepts Claude already knows, but the air-hatch rationale ("can be air-hatched with a clear \"SKIPPED\" status that the reviewer will see and act on") and the SKIPPED-status explanation are repeated across Steps 2, 3, 6, and 7 and could be trimmed. Fits "efficient; minor instances of over-explanation" (4), not the lean 5. | 4 / 5 |
Actionability | Every step ships copy-paste-ready bash (git ls-files, sort -V | tail -1, stat -f %m / stat -c %Y), exact match regexes ("^- Must Fix open: (\\d+)$"), verbatim failure and confirmation messages, and complete stub templates. This matches the fully-executable anchor, and the few placeholders (<missing files>, <latest-code-review-report-path>) are deliberate runtime substitutions. | 5 / 5 |
Workflow Clarity | Steps 0–8 are explicitly sequenced, each verification step produces a PASS/FAIL/SKIPPED verdict, the submit is gated ("If any check is FAIL: stop here. Do not run the CLI"), a summary table and explicit confirmation precede the destructive submit, and every failure message names the recovery skill to re-run. This is a full feedback-loop workflow matching the 5 anchor, well above the validation-capped 3. | 5 / 5 |
Progressive Disclosure | A single-file skill (no references/, scripts/, or assets/ exist) with clear section structure; the stub templates are one level deep and correctly signaled from Step 0 ("the templates in 'Stub templates' at the end of this skill"). Minor gap: the ~40-line stub-template section plus the Step 0 air-hatch branching could live in a reference file to slim the always-loaded body, so it fits 4 rather than 5. | 4 / 5 |
Total | 18 / 20 Passed |