Content
46%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 well-organized catalog with real safety constraints (RoE intersection, non-destructive proofs, forbidden-destination checks), but it front-loads conceptual explanation Claude already knows, never sequences the emulation workflow with explicit validation checkpoints, and — most significantly — completely fails to link its own two reference files, leaving the IOC and CVE-APT mapping content undiscoverable.
Suggestions
Add a "References" section linking the bundle files, e.g., "**IOC lookup**: See [references/anyrun-free-iocs.md](references/anyrun-free-iocs.md)" and "**CVE-to-actor mapping**: See [references/cve-apt-mapping.md](references/cve-apt-mapping.md)", so the existing reference material is discoverable.
Replace the conceptual opening (definition of emulation vs. simulation) with a numbered workflow: select profile → intersect techniques with RoE → substitute non-destructive proofs for Impact techniques → execute with forbidden-destination checks → classify detection evidence, with an explicit validation gate before any destructive-adjacent step.
Trim the introductory paragraph and the "When to use this" section (which duplicates the frontmatter when_to_use) to the one or two sentences that add non-obvious information.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The opening paragraph spends ~10 lines defining adversary emulation and distinguishing it from pentesting and simulation — concepts Claude already knows — and the "When to use this" section duplicates the frontmatter when_to_use field. This is more than the 'minor instances of over-explanation' of level 4, but the rest (catalog table, interpretation rules) is dense and earns above level 2. | 3 / 5 |
Actionability | There is concrete guidance — the `load_skill <slug>` catalog, non-destructive proof substitutions ("a benign canary file, a dry-run"), "Intersect each selected technique with approved RoE", and the ATT&CK groups verification URL — but no complete executable procedure for building or running an emulation plan. This sits between level 2's high-level-only hints and level 4's mostly-executable guidance. | 3 / 5 |
Workflow Clarity | Profile selection → RoE intersection → execution → detection evidence is only loosely implied, never sequenced, and validation checkpoints ("Out-of-scope techniques remain forbidden", "Forbidden-destination checks still apply at tool time") are implicit rather than explicit steps. Because the subject matter includes destructive-capable techniques emulated without an explicit validate-before-proceeding workflow, workflow clarity is capped at 3. | 3 / 5 |
Progressive Disclosure | The bundle contains two substantial reference files (references/anyrun-free-iocs.md, references/cve-apt-mapping.md) that are never mentioned, linked, or signaled anywhere in the SKILL.md body — they are undiscoverable from the index, which matches level 2's 'references are buried' condition. The body's own section structure is reasonable, keeping it above level 1, but the orphaned bundle references prevent level 3+. | 2 / 5 |
Total | 11 / 20 Passed |