Content
92%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-engineered instruction skill: fully actionable with concrete commands, API endpoints, and templates; a clearly sequenced multi-stage workflow with explicit confirmation/validation gates and error-recovery notes; and clean progressive disclosure through three real, one-level-deep references. Only minor conciseness trim opportunities keep it from a perfect score.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean imperative/bullet prose that assumes Claude's competence and avoids explaining basic concepts, but a few rationale sentences (e.g. on flaws and on the identity.md pitfall) could be trimmed slightly. Not a full 5 because there is minor explanatory padding that earns less than every token. | 4 / 5 |
Actionability | Provides copy-paste-ready, executable guidance throughout: concrete commands (`web_search site:clawhub.ai`, `install_skill --github_url`, `ls ~/.hanako/skills`, `media_get-guide`, `current_status` session_files), real API endpoints (`POST /api/character-cards/plan` and `/import`), exact file paths, and inline markdown templates for identity.md and AGENTS.md. | 5 / 5 |
Workflow Clarity | Seven stages (阶段〇–七) are explicitly sequenced with validation/confirmation gates — 复述确认方向, user confirm before proceeding, and the plan-preview → user-confirm → import feedback loop — plus a 常见状况 section covering error recovery (中途改主意, id撞车, 只要zip). The batch/import operation has explicit validation, so the destructive-cap does not apply. | 5 / 5 |
Progressive Disclosure | Clear overview structure with three well-signaled one-level-deep references ("先读/必读 references/character-craft.md", "references/anti-slop.md", "references/card-format.md"), all confirmed real files; methodology and format detail are appropriately split out while templates stay inline. | 5 / 5 |
Total | 19 / 20 Passed |