Content
88%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 operational skill: nearly every line is a specific, verifiable instruction or footgun Claude would not know on its own, with executable code throughout and validation woven into every destructive workflow. The main improvement opportunities are splitting the design-guidance section into a referenced file and trimming a few advisory passages.
Suggestions
Move the Design Ideas section (palettes, layout options, typography tables) into a references/design.md and keep a short decision-oriented summary inline, so editing/reading tasks don't load ~80 lines they never use.
Tighten the pptxgenjs gotchas list by grouping related footguns (e.g., merge the file-corrupting color/shadow/offset items into one 'corrupts the file' cluster) to cut redundant framing sentences.
The QA visual-check bullet list partially duplicates the Avoid list in Design Ideas; consolidate into a single checklist to remove near-duplicate guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, hard-won operational knowledge ('Hex colors: never #, never 8 digits... Both "#FF0000" and alpha baked into the hex corrupt the file', 'letterSpacing is silently ignored — the real option is charSpacing') and assumes Claude's competence, with essentially no padding explaining known concepts. Minor trimming is possible in the Design Ideas and QA prose (e.g., 'After staring at the generating code you tend to see what you expect rather than what rendered'), keeping it just below the 'every token earns its place' anchor. | 4 / 5 |
Actionability | Guidance is fully executable and copy-paste ready: complete bash blocks for unpack/edit/pack/validate ('cd unpacked && rm -f ../out.pptx && zip -Xr ../out.pptx .'), exact pptxgenjs option names and values (dataLabelPosition: ctr/inEnd/inBase, valAxes + catAxes two entries each), a concrete placeholder-detection grep, and full PDF-to-image commands. It covers all common cases across create, edit, read, and QA. | 5 / 5 |
Workflow Clarity | Workflows are clearly sequenced with explicit validation checkpoints and feedback loops despite destructive/batch operations: 'After writeFile(), run python scripts/office/validate.py deck.pptx... Fix them in your generator, not by hand-editing', 'Do all structural work — add, delete, reorder — before editing any slide's content' (with the reason each ordering matters), '--original baselines the schema checks against the template', and 'After fixes, rerun all four commands above'. The required QA section (content, file, visual) closes the loop with re-render-and-recheck guidance. | 5 / 5 |
Progressive Disclosure | Structure is good: a task-dispatch table up top ('Create → pptxgenjs script; Edit → unzip → edit slideN.xml → zip; Read → markitdown'), a scripts table where every referenced path (scripts/thumbnail.py, add_slide.py, clean.py, office/validate.py, office/soffice.py) is a real bundle file, one internal anchor link, and no nested references at all. The gap keeping it from 5: everything is inline in a 238-line SKILL.md — the ~80-line Design Ideas section (palettes, layouts, typography) is prime material for a one-level-deep references/ file that only creation tasks need to load. | 4 / 5 |
Total | 18 / 20 Passed |