Content
75%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-structured, lean workflow that tells Claude exactly what to search, how to grade overlap risk, and what report to produce, with a sensible deferral of detail to a shared reference. Main gaps are a few non-executable placeholder steps (espacenet query construction, classification searching) and the absence of explicit validation/retry checkpoints in the search loop.
Suggestions
Add a concrete, executable espacenet pattern (e.g. an example worldwide.espacenet.com query URL with `q=`, `Classification=` parameters) so Step 2 is copy-paste ready.
Add an explicit validation checkpoint, e.g. 'if a WebFetch fails or returns no patent details, retry with an alternate keyword set before dropping the reference' and 'cap detailed analysis at MAX_PATENT_RESULTS, listing the rest as titles only'.
Trim non-operational lines like 'Adapted from `/research-lit` for patent-specific searching.' to save tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient — constants, a 6-step workflow, and a copy-ready report template with no padding or explanations of concepts Claude already knows (it never explains what a patent or prior art is). It is not a 5 because of minor trimmable content: the 'Adapted from `/research-lit`' provenance line and some verbose phrasing in Step 5's disclaimer add little operational value. | 4 / 5 |
Actionability | Mostly executable guidance: concrete query patterns ("site:patents.google.com [keywords]", "[assignee name] patent [technical area]"), a shared reference for search templates, and a fully specified output report with table schemas. It is not a 5 because several steps remain template placeholders ("WebFetch worldwide.espacenet.com/search results" gives no concrete query-URL pattern, and 'search by predicted IPC/CPC classes' lacks a worked example), and not a 3 because the given commands and output structure are directly usable. | 4 / 5 |
Workflow Clarity | A clearly sequenced 6-step workflow (extract concepts → patent search → literature search → classify/analyze → FTO assessment → output) with defined inputs, constants, and a 'Never fabricate... Mark uncertain references with [VERIFY]' integrity rule acting as a checkpoint. It is not a 5 because there is no explicit validate-and-retry loop (e.g. re-checking a fetched patent page that fails to load, or capping results against MAX_PATENT_RESULTS). This is a search/report task, not a destructive or batch-mutating operation, so the workflow-clarity cap of 3 does not apply. | 4 / 5 |
Progressive Disclosure | Good structure: sections are organized (Constants, Inputs, Shared References, Workflow, Key Rules), the one external reference (`../shared-references/prior-art-databases.md`) is clearly signaled and one level deep, and the bulk detail (search strategy templates, IPC/CPC guidance) is correctly deferred to it. It is not a 5 because no bundle files exist in the skill directory to verify against and the body still inlines the full report template plus several sub-strategies that could live in the shared reference; it is not a 3 because what is inlined is appropriately concise and the reference is well signaled. | 4 / 5 |
Total | 16 / 20 Passed |