Content
80%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The content is concise and highly actionable with concrete patent-specific templates and a reviewer prompt, but the workflow's refinement loop is declared as a constant yet never operationalized as an explicit validate-revise-retry cycle, and its one external reference points to a file that does not exist.
Suggestions
Make the refinement loop explicit in the workflow: after Step 6, add 'If the reviewer identifies issues, revise the flagged layer (Step 1/2/4) and re-run validation, up to MAX_REFINEMENT_ROUNDS; only proceed to Step 7 when no major issues remain.'
Either create ../shared-references/patent-writing-principles.md with the referenced Problem-Solution-Advantage framework, or inline the essential parts and drop the dangling reference so the body is self-contained.
Move the claimable-subject-matter and drawing-plan tables into a references/ file if they grow, keeping SKILL.md an overview that links one level deep — this would lift progressive_disclosure toward the score-3 anchor.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body gives domain-specific patent guidance (Problem-Solution-Advantage framing, claim-category and drawing-plan tables, a concrete reviewer prompt and output template) without explaining concepts Claude already knows, and is structured rather than padded — matching the lean, score-3 anchor; only minor redundancy between Step 1 and Key Rules keeps it from being flawless. | 3 / 3 |
Actionability | Provides concrete executable guidance: a fixed problem-statement format, claimable-subject-matter and drawing-plan tables, a copy-paste mcp__codex__codex prompt with explicit config, and a complete output template — fully actionable for an instruction-only skill, matching the score-3 anchor. | 3 / 3 |
Workflow Clarity | The 7 steps are clearly sequenced and Step 6 is a validation checkpoint, but the feedback loop is only implicit: MAX_REFINEMENT_ROUNDS=3 is declared yet the workflow never states what to do with reviewer findings (revise which step, re-run, stop) — matching the score-2 anchor of steps present with checkpoints missing or implicit. | 2 / 3 |
Progressive Disclosure | Sections are well-organized, but all content is inline (~180 lines) and the single external reference '../shared-references/patent-writing-principles.md' does not exist on disk (no bundle files present; /research-refine also missing), so the reference is not a real navigable path — matching the score-2 anchor of structure present but references not clearly signaled / content that could be separate is inline. | 2 / 3 |
Total | 10 / 12 Passed |