Content
85%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-structured instruction-only skill: the workflow is clearly sequenced with real validation checkpoints, and progressive disclosure is exemplary — a lean overview pointing to four well-described, one-level-deep reference files. The main costs are cross-section repetition of a few core rules and the absence of any inline example fragment to ground the abstract seven-part framework.
Suggestions
State the never-merge rule once (e.g., in Principles) and have the Delivery section and Failure patterns reference it instead of restating it, trimming roughly a paragraph of repetition.
Inline one short filled example of a header and one Gate 0 check line (2-4 lines total) so the seven-part framework is grounded without opening references.
Consider narrowing the broadest trigger phrases in the description ("background agent", "run this overnight") to dispatch-specific phrasings to reduce overlap with agent-orchestration skills.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence — it never explains what a PR, worktree, or verification is — and every section adds non-obvious guidance. However, core rules are restated across sections: the never-merge rule appears three times ("The agent never merges", "It never merges", "Permission to merge"), and the derive-counts and one-worktree rules each appear twice in Principles and Failure patterns. This is efficient with minor trimmable repetition, fitting anchor 4 rather than the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | Guidance is concrete and executable for an authoring task: a 9-step workflow, an exact output filename convention ("dispatch-[short-name]-[YYYY-MM-DD].md"), and specific directives like "Point at the source spec instead of pasting it" and "Ask for the output to be pasted, not summarized". It stops short of anchor 5 because the body shows no inline fragment of any dispatch part (header, Gate 0 check) — all examples are deferred to the reference files — leaving minor gaps if the references are not opened. | 4 / 5 |
Workflow Clarity | The 9-step workflow is clearly sequenced with explicit validation checkpoints: step 1 gates readiness ("If you do not know how the work will be verified, stop and find out") and step 8 is a review-as-the-agent feedback loop ("Find every place you would have to ask a question. Answer it... or turn it into a stop condition"), reinforced by the Gate 0 checklist reference. This matches anchor 5's clear sequence, explicit validation, and feedback loops for error recovery. | 5 / 5 |
Progressive Disclosure | The body is a genuine overview (framework, workflow, failure patterns) with all detail split into four reference files that each carry a one-line description of when to use them, linked both inline at the point of use in the workflow and again in a dedicated Reference files section. All four referenced files exist and contain no nested references, so navigation is one level deep and easy — a clear anchor-5 match. (The unreferenced _claude-frontmatter-extras.yaml is a porting artifact, not skill content.) | 5 / 5 |
Total | 18 / 20 Passed |