Content
81%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 unusually rigorous instruction-only skill: fully actionable templates, a validation-gated workflow with repair loops, and a real one-level-deep reference for report format. Its main cost is token weight — core rules are repeated three to four times across the rules, drafting-prompt, and acceptance sections, which could be consolidated without losing any constraint.
Suggestions
State each repeated rule once: the "witness" ban, actor-naming convention, absolute-path ban, and report-only conditionality each appear in "Non-negotiable rules", the drafting prompt, and "Report acceptance"; consolidate them into a single authoritative section and have the acceptance checklist reference the rule names instead of restating them.
Deduplicate the release-history guidance that appears in both "Trace affected release history" and the drafting prompt's third paragraph, keeping the full procedure in one place and a one-line summary in the other.
Trim the overlap between "Source and exploitability standard" and the drafting prompt's causal-order and fix-verification paragraphs, which restate the same requirements in near-identical wording.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every rule is skill-specific policy rather than concepts Claude already knows, but the same rules are restated across sections — the "witness" ban (lines 27, 138, 178), actor naming (lines 31, 36, 136, 179), the absolute-path ban (lines 32, 116, 142, 182), and report-only conditionality (lines 23, 128, 144, 169) — so the body could be tightened by roughly a third. It is mostly efficient but noticeably redundant between "Non-negotiable rules", the drafting prompt, and "Report acceptance". | 3 / 5 |
Actionability | The skill supplies a copy-paste-ready fill-in-the-blank single-finding drafting prompt, a rewrite prompt template for failed drafts, exact output paths (`findings/<slug>/<slug>.md`), a concrete git command (`git show REV:PATH`), fixed actor/username conventions (alice/bob/mallory/eve), worked phrasing examples, and explicit stall and delegation-failure protocols. | 5 / 5 |
Workflow Clarity | The 10-step campaign workflow is clearly sequenced with explicit validation checkpoints (step 7 independent re-reading against artefacts, step 8 enumerated reject criteria), a repair feedback loop (step 9: fresh sub-agent with specific failures), a full acceptance checklist, and defined handling for unavailable delegation and stalled workers. | 5 / 5 |
Progressive Disclosure | Report-format detail is properly externalized to the clearly-signaled, one-level-deep `references/report-format.md`, which exists in the bundle. However, the ~184-line body still inlines the complete drafting-prompt template and long policy sections that could be split out, and the `../../references/artifact-storage.md` pointer resolves outside the bundle. | 4 / 5 |
Total | 17 / 20 Passed |