Content
88%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 concise, well-sequenced operating guide with strong validation and rollback discipline appropriate to destructive ad-account mutations. Its main weakness is progressive disclosure: it relies on external shared files rather than a verifiable in-skill bundle, and a few directives are high-level rather than concrete.
Suggestions
Add a brief in-skill reference (e.g. references/operating-contract-summary.md) or inline the essential operating-contract rules so the skill is self-contained when the shared file is absent.
Tighten the few high-level directives into concrete steps — for example, specify which read capability retrieves lead responses and how to minimize PII exposure in the response.
Confirm the linked shared paths (../shared/*.md, ../../docs/mcp-connection.md) resolve within the deployed skill tree, or move the MCP connection steps inline to avoid broken navigation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean imperative prose with no padding or re-explanation of concepts Claude knows; tokens are spent on platform-specific facts Claude would not reliably know (major-unit money objects, immutability, URN resolution, hashing boundaries), matching the lean-and-efficient anchor; not a 4 because there is no over-explanation to trim. | 5 / 5 |
Actionability | Concrete, specific guidance — 'Use dedicated write tools, never the read-only script surface', 'Resolve targeting names to LinkedIn URNs before setting the full targeting tree' — with minor gaps in fully executable detail; not a 5 because some directives stay high-level ('Retrieve only what the question needs') rather than giving copy-ready specifics, and not a 3 because the guidance is largely executable and platform-specific. | 4 / 5 |
Workflow Clarity | Sequenced numbered steps with explicit validation checkpoints and feedback loops — dry-run previews, draft-then-review-then-activate, 'Verify the mutation through returned before/after evidence', rollback triggers — so the destructive-operation cap does not apply; not a 4 because validation and error-recovery are explicit throughout. | 5 / 5 |
Progressive Disclosure | Well-sectioned overview that points to shared materials via clearly signaled links (operating-contract.md, measurement-framework.md, mcp-connection.md) at one level deep, but no in-skill bundle files exist and the referenced shared paths are external/unverified; not a 5 because navigation depends on out-of-bundle files whose presence cannot be confirmed, and not a 3 because structure and signaling are otherwise good. | 4 / 5 |
Total | 18 / 20 Passed |