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.
A lean, highly actionable skill body with a well-sequenced workflow, complete schema documentation, and a strong RoE-validation checkpoint. Its main weaknesses are the disconnect between the body's navigation and the actual bundle (missing emulation/ paths, unlinked references files) and the absence of an output-verification feedback loop.
Suggestions
Link the bundled reference files from the relevant steps — e.g., in Step 2 point to references/adversary-archetypes.md for pre-built tier profiles and in Step 5 to references/apt-groups.md for named-actor TTP data — and reconcile the 'emulation/*/SKILL.md' paths with what actually ships in the bundle.
Add an explicit output-validation step after Step 4: verify plan/threat-profile.json against the ThreatProfile schema (e.g., run the validator) and fix-and-retry on failure, rather than only stating the document 'must validate'.
Trim the overlap between Step 2's playbook-loading parenthetical and Step 5, folding the load_skill mechanics into one place.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — tier tables, field lists, RoE constraint tables, and a complete JSON example with no explanations of what APT groups or MITRE ATT&CK are. Minor trimming is possible: the Step 2 parenthetical about loading emulation playbooks and 'Skillogy returns skill bodies, not sibling reference files' overlaps with Step 5's fuller treatment, and the rhetorical opener ('devolves into arbitrary tool usage') is slight padding — this matches 'Efficient; minor instances of over-explanation that could be trimmed' rather than the perfectly lean anchor 5. | 4 / 5 |
Actionability | Fully executable guidance throughout: a complete copy-paste 'plan/threat-profile.json' example with realistic values, exact StrEnum values ('tier-1'..'tier-4'), exact load_skill paths, the legacy embedded-summary shape with which fields to skip, an explicit RoE-to-technique mapping table, and a full field/type/required/allowed-values schema table. Specific examples cover the common cases (standalone and embedded output), matching the top anchor. | 5 / 5 |
Workflow Clarity | A clear 5-step sequence (tier → profile → RoE validation → output → optional kill chain) with an explicit validation checkpoint in Step 3 (profile must be consistent with RoE constraints) and a pruning instruction in Step 5 ('Prune any technique the RoE forbids (Step 3) before writing'). It falls short of anchor 5 because there is no verification of the produced output against the ThreatProfile schema — no validate/fix/retry feedback loop after writing — matching 'most checkpoints present; minor validation gaps'. This is a document-authoring workflow, so the destructive/batch cap at 3 does not apply. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned and points to detail files via load_skill, but scored against the actual bundle: the referenced 'emulation/SKILL.md' and 'emulation/<actor>/SKILL.md' paths do not exist in the bundle, while the two files that do exist (references/apt-groups.md with per-group TTP data and references/adversary-archetypes.md with pre-built tier profiles — exactly the content Steps 2 and 5 need) are never linked or signaled from the body. This orphaned/mismatched reference structure matches 'references present but not clearly signaled; could be better organized' (anchor 3) rather than anchor 4's 'references mostly clear'. | 3 / 5 |
Total | 16 / 20 Passed |