Content
78%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.
The body is a lean, well-structured governance skill: concrete rules, a sequenced workflow with evidence-based checkpoints, and proper delegation of detail to a single clearly-labeled reference. Its main weakness is undefined project-specific jargon ("native RDD consent", "chain strategy/exception", "Gentle AI" references) that a portable skill should define or drop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no basic-concept padding — and every section carries rules ("never substitute Gentle AI's rules", "Never stash/pop for this purpose"). Minor repetition keeps it from a 5: label-mutation authority is stated three times ("Never invent a label...", "never infer mutation authority from a discovered label", "Apply labels only through the canonical workflow...") and authorization rules recur across Hard Rules and Execution Steps. Not 3 because nothing explains concepts Claude already knows and the ~40 lines could not be cut substantially without losing rules. | 4 / 5 |
Actionability | Most guidance is concrete and executable: "Read the destination's `CONTRIBUTING.md`, agent instructions, PR template", "Closes/Fixes/Resolves #N" vs "Refs #N", "`viewerPermission` `MAINTAIN` or `ADMIN`", and explicit prohibitions like "Never stash/pop". Not 5 because several terms are never defined — "native RDD consent", "chain strategy/exception", "Gentle AI's Go/Docker examples" — leaving those instructions non-executable without external context. Not 3 because the bulk of the guidance names exact files, permission levels, and actions. | 4 / 5 |
Workflow Clarity | Execution Steps 1–5 give a clear sequence with explicit checkpoints: "run the project's applicable checks", baseline verification ("the same failing command/environment on a comparable isolated clean base"), "Mark checkboxes only after observed evidence", and "one bounded attempt and target-host readback". Not 5 because there is no fix-and-retry feedback loop — failures are reported ("Report failures honestly") rather than iterated — and the Decision Gates table conditions are not all resolvable procedurally. Not 3 because validation checkpoints are explicit at nearly every step. | 4 / 5 |
Progressive Disclosure | The body is under 50 lines with well-organized sections (Activation Contract, Hard Rules, a Decision Gates table, Execution Steps, Output Contract) and a single clearly-signaled, one-level-deep reference: "`../issue-creation/SKILL.md` — canonical action authority", which appropriately delegates the detailed label-mutation workflow. No bundle files exist to inline, and nothing that belongs in a separate file is inlined. Not 4 because there are no organization gaps — the split between overview and delegated detail is exactly right for a lean skill. | 5 / 5 |
Total | 17 / 20 Passed |