Content
82%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-structured, highly actionable instruction skill: concrete templates, decision matrices, verdict criteria, and an explicit independent-verification step with a fallback path. Weaknesses are minor — reviewer-routing boilerplate padding, no handling for a missing prior-art input, and unverifiable out-of-bundle shared references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and task-focused with no explanation of concepts Claude already knows (no patent-law primer, no library overviews); tables, templates, and rules are all load-bearing. The main trimmable content is the reviewer-routing boilerplate block and the "Adapted from /novelty-check" lineage note, which are minor instances of over-explanation — matching the 'efficient; minor instances that could be trimmed' anchor rather than the every-token-earns-its-place level 5. | 4 / 5 |
Actionability | Guidance is fully executable for an instruction-only skill: exact anticipation matrix layout with verdict rules ("ANTICIPATED: One reference discloses every element"), a complete spawn_agent template with model and reasoning_effort specified, defined motivation-to-combine criteria, per-jurisdiction pass/fail formats, and a copy-paste-ready output file template. It is not level 4's 'minor gaps' — the common cases (novel, obvious, needs amendment) each have concrete templates covering them. | 5 / 5 |
Workflow Clarity | Six clearly sequenced steps with an explicit independent-examiner verification checkpoint (Step 4) and a rejected-claim feedback path (recommended amendments), plus an explicit fallback when reviewer delegation is unavailable. It falls short of level 5 because some checkpoints are missing or implicit: there is no instruction for what to do when the required input patent/PRIOR_ART_REPORT.md does not exist, and no confirmation that the output file was written correctly. It is clearly above level 3, which lacks checkpoints entirely. | 4 / 5 |
Progressive Disclosure | The single-file body is well-sectioned (Constants, Inputs, Shared References, Workflow, Key Rules) and shared references are clearly signaled in a dedicated section pointing one level deep (patent-writing-principles.md, patent-format-us.md, reviewer-backends.md). It sits at the 'good structure, references mostly clear' level rather than 5 because the shared references live outside the skill bundle and cannot be verified as navigable from it, and some framework detail (jurisdiction standards) is inlined that could arguably live in those references. | 4 / 5 |
Total | 17 / 20 Passed |