Content
100%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 exemplary instruction-only skill body: concrete executable commands, a tightly sequenced workflow with validation and rollback checkpoints at every risky step, and disciplined token efficiency. The untrusted-PR-metadata note is a justified, behavior-changing addition rather than padding.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and command-first: every line carries non-obvious workflow detail (marker-commit cutoff with 1-week overlap margin, ff-only pulls, staged-only guard) and nothing explains concepts Claude already knows. No padding is detectable to trim. | 5 / 5 |
Actionability | Guidance is fully executable and copy-paste ready, from the worktree-locating pipeline and the gh pr list command with pushed-down search filter and --jq projection, to the exact commit message with the [contributors-updated] marker and the test-based staged-change verification. | 5 / 5 |
Workflow Clarity | Six clearly sequenced steps with explicit validation checkpoints and feedback loops throughout: stop on dirty tree, user approval gate before any edit, re-pull with --ff-only immediately before commit, staged-file equality check, git diff --cached --check, and a stop/recompute/re-approve path if master moved. The batch/direct-push validation cap is satisfied, not violated. | 5 / 5 |
Progressive Disclosure | The skill is a single linear workflow (~100 lines) with well-organized numbered sections and no content that belongs in separate files; there are no bundle references to verify, and nothing is inlined that should be split out. | 5 / 5 |
Total | 20 / 20 Passed |