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 well-organized diagnostic skill: layer-based framework, concrete pattern-to-cause tables, explicit validation checkpoints, and a properly disclosed one-level-deep reference checklist. The main gaps are the missing example tool invocations (which Ahrefs MCP / Search Console call to make for each pull) and an explicit loop-back instruction when the root-cause hypothesis is refuted.
Suggestions
Add one example invocation per layer (or in the checklist reference), e.g., the specific Ahrefs MCP tool name and a sample Search Console query/filter for the Layer 3 position-history pull, to lift actionability to fully executable.
Make the refutation loop explicit in the Workflow: state what to do when step 8's hypothesis is refuted (e.g., 'return to the deepest layer that produced the hypothesis and re-segment with the next pattern from the Layer 2 table').
Trim the few near-tautological lines (e.g., 'A change in one segment requires different diagnosis than a change everywhere') and let the Workflow reference layers by what to do rather than restating their titles.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with domain-specific diagnostic knowledge Claude does not already have (segmentation patterns, pattern-to-cause table, failure modes like 'premature reassurance'), with almost no padding or explanation of known concepts. It is not 5 because of minor trim opportunities, e.g. 'A change in one segment requires different diagnosis than a change everywhere' and the Workflow section restating the layer names already detailed above; not 3 because those instances are minor and every major section earns its tokens. | 4 / 5 |
Actionability | As an instruction-only skill the absence of code is not penalized, and the guidance is concrete: exact segments to pull (country, device, branded vs non-branded), a pattern-to-likely-cause lookup table, specific technical checks (robots.txt, canonical tags, accidental noindex, hreflang, redirect chains), and a referenced per-layer checklist of pulls and checks. It is not 5 because the body (and checklist) name data sources but never show an example invocation — e.g., which Ahrefs MCP tool call or Search Console filter to use for a given pull; not 3 because the guidance is directly executable at the checklist level with only minor gaps. | 4 / 5 |
Workflow Clarity | The 10-step Workflow is clearly sequenced and mapped to the five layers, with explicit validation checkpoints ('Validate the data. Layer 1 checks. Rule out tracking and seasonality' and 'Validate the hypothesis. Find the evidence that confirms or refutes it') plus a Failure patterns section that serves as error-recovery guidance. It is not 5 because the refutation path is implicit — step 8 says the hypothesis may be refuted but never states where to loop back to (which layer to revisit); not 3 because this is a read-only diagnostic skill with no destructive/batch operations and multiple checkpoints are present. | 4 / 5 |
Progressive Disclosure | The body is an overview/framework and the per-layer operational detail (specific data pulls and per-signal interpretation) is appropriately delegated to references/diagnosis-checklist.md, which exists, is one level deep, and is clearly signaled twice (Workflow step 8 with link, and a dedicated Reference files section with a one-line description). It is not 4 because the split is clean and navigation is easy — no content that clearly belongs in the reference is inlined, and there is no nesting beyond one level. | 5 / 5 |
Total | 17 / 20 Passed |