Content
77%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.
A well-engineered process skill: explicit step sequence with real validation gates and destructive-operation safeguards, copy-ready git/gh commands, and clean two-file progressive disclosure. Its one real weakness is token efficiency — the E7 memory-path essay and the Hard-nevers section restate the philosophy section at noticeable length.
Suggestions
Compress E7 to its actionable rule (never write memory autonomously; capture human-recognized patterns with them and commit on the feature branch) and cut the /learn-P7 ground-truth mechanics, which belong in the /learn skill, not here.
Replace the full-sentence Hard nevers that merely restate E3/E7/E8/E9 with one-line pointers (e.g. "Never rehash bot findings (E3)"), keeping the list as a compact checklist.
Add a concrete command for finding the handoff PR in Step 0 (e.g. a gh pr list/--search invocation for the "Ready for your review" comment) so every step is executable as written.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nothing explains concepts Claude already knows, but E7 ("Memory written here is the human's call, and it's authoritative...") sprawls across two long paragraphs of /learn mechanics, and "Hard nevers" restates E3, E7, E8, and E9 nearly verbatim with cross-references. The duplication is more than the minor trimming anchor 4 allows; fits anchor 3's "could be tightened". | 3 / 5 |
Actionability | Concrete, executable guidance throughout: "git checkout --detach origin/feature/<f>", "gh pr diff <pr>", "git push origin HEAD:feature/<f>", plus exact file paths ("prds/<f>/prd.md", "run-prd-test.sh"). Minor gaps — "find the open PR carrying the harness's 'Ready for your review' handoff comment" and the merge/close mechanics have no command (delegated to verdict.md). Anchor 4. | 4 / 5 |
Workflow Clarity | Steps 0–8 are clearly sequenced with explicit validation checkpoints: clean-tree preflight, the Step 6 gate ("Do you feel you understand this change?"), explicit human go-ahead required for the destructive merge/close, a fix-push-review feedback loop ("once it's clean again you merge"), idempotent re-run handling, and a Hard-nevers checklist. Matches anchor 5. | 5 / 5 |
Progressive Disclosure | The body is an overview-plus-guided-flow that splits depth into two real, one-level-deep references ("references/walkthrough.md", "references/verdict.md"), each clearly signaled with a content description and hackable-seam annotation; the reference files themselves nest no further. Matches anchor 5's well-signaled, easy-navigation structure. | 5 / 5 |
Total | 17 / 20 Passed |