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-engineered orchestration skill: workflows are explicitly sequenced with validation gates and resume logic, detail is properly pushed to real, well-signaled one-level-deep reference files, and scripts named in the body all exist in the bundle. The only minor gaps are occasional trimmable rationale and command invocations that live in references rather than inline.
Suggestions
Add a brief quick-reference of exact one-line script invocations (fork.sh, sync-patches.sh, select-mutant.sh, verify-mutant.sh) so the most common steps are copy-paste ready without opening references/mutation-harness.md.
Tighten justification sentences (e.g. the baseline-gate and Sometimes-class rationale) to a rule plus a single-clause reason, trusting the reader with the why behind each gate.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, high-signal, and skill-specific — no general-concept padding — but justification prose such as "Mutation testing on a buggy SUT measures nothing: you cannot tell a property that fired because of your mutant from one that was already failing" and the layered qualifier sentences around scoping could occasionally be trimmed to a rule plus a one-line reason. | 4 / 5 |
Actionability | Guidance names concrete scripts and flags ("verify each mutant locally with `verify-mutant.sh`", "export them with `sync-patches.sh`", "launched with a dedicated `--source` and `--ephemeral`") plus exact artifact paths, but exact command invocations are deferred to reference files rather than being copy-paste ready inline. | 4 / 5 |
Workflow Clarity | Three explicitly sequenced workflows (First sweep, Iterate after a sweep, Resume a stopped sweep) with strong validation checkpoints — the baseline gate, local verification before spending run budget, the two-branch re-run rule — and a Self-Review checklist providing feedback loops, matching the exemplar anchor. | 5 / 5 |
Progressive Disclosure | Clear section structure with a Reference Files table mapping each of the seven references (all present on disk) to a when-to-read condition, one-level-deep references cited with section titles (e.g. "references/sweep-and-verdicts.md, 'Attributing a falsification'"), and every cited script present in assets/mutation-testing/. | 5 / 5 |
Total | 18 / 20 Passed |