Content
77%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 strong, dense skill body with excellent verification discipline (explicit fix-revalidate loop, subagent prompt, placeholder grep) and fully executable read/render commands. Its main defects are the two dangling reference files — the editing and from-scratch creation guides are pointed to but missing from the bundle, leaving the primary workflows without executable detail — and some bulk inline design content that could be externalized or trimmed.
Suggestions
progressive_disclosure: Add the missing editing.md and pptxgenjs.md files to the bundle (e.g., under references/) — or, if they cannot be provided, inline the essential unpack/edit/pack steps so the two primary workflows are self-contained.
actionability: Include the concrete commands for the editing workflow in the body (unpack, the slide-editing step, clean, pack via scripts/office/) so the '解包 → ... → 打包' arrow sequence is executable without leaving SKILL.md.
conciseness: Consolidate the repeated visual-element rules in the '避免' section with the '每张幻灯片' guidance above it, and consider moving the 10-row color palette and font pairing tables to a reference file to cut inline tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and opinionated rather than padded — it explains nothing Claude already knows, and the tables (color palettes, font pairings, sizes) are non-obvious domain specifics that earn their tokens. However, there are minor instances of over-explanation: repeated exhortations in the '避免' section restate points already made ('不要创建纯文本幻灯片' repeats the '每张幻灯片都需要视觉元素' rule), and the 10-row palette table plus font tables are bulk inline content that could be trimmed or externalized. This fits 'Efficient; minor instances of over-explanation that could be trimmed' rather than the lean anchor 5. | 4 / 5 |
Actionability | Reading, rendering, and verification are fully copy-paste ready: 'python -m markitdown presentation.pptx', 'python scripts/thumbnail.py presentation.pptx', 'python scripts/office/soffice.py --headless --convert-to pdf output.pptx', 'pdftoppm -jpeg -r 150 output.pdf slide', the placeholder grep, and a complete subagent prompt template. The gap keeping it below 5 is that the two primary workflows defer all executable detail — the editing workflow is only a high-level arrow sequence ('解包 → 操作幻灯片 → 编辑内容 → 清理 → 打包') with no commands, and the creation workflow is a bare pointer — and their targets (editing.md / pptxgenjs.md) are absent from the bundle, so the deferred guidance is unreachable. This fits 'Mostly executable guidance; concrete code or commands with minor gaps' rather than the incomplete/pseudocode anchor 3. | 4 / 5 |
Workflow Clarity | The verification loop is exactly the anchor-5 pattern: generate → convert to images → check (with both a content check command and a visual-check subagent prompt) → list problems → fix → '重新验证受影响的幻灯片' → repeat until a full pass finds nothing new, plus the explicit gate '在完成至少一个修复-验证循环之前,不要宣布成功'. Checklists are provided (the 12-item visual-check list, the placeholder grep). Editing/packing validation is also present via scripts/office/validate.py in the bundle, and the workflow for the destructive unpack/edit/pack cycle has clear checkpoints. It is not below 5 because no validation gaps exist in the written workflows. | 5 / 5 |
Progressive Disclosure | Structure and signaling are good — a quick-reference table routes each task to the right guide, references are one level deep, and scripts referenced in the body (thumbnail.py, office/unpack.py, office/soffice.py) all exist in the bundle. However, the two main workflow guides the body links to, [editing.md](editing.md) and [pptxgenjs.md](pptxgenjs.md), do not exist anywhere in the bundle (no references/ directory), so the primary navigation paths are broken. That lands at 'Some structure but could be better organized; references present but not clearly signaled' territory — the organization is better than that anchor, but the dangling references prevent the good-structure (4) or clear-overview (5) anchors, and a large design section sits fully inline. | 3 / 5 |
Total | 16 / 20 Passed |