Content
63%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.
The body is a well-structured, highly actionable overview for reading, converting, and QA-ing pptx files, with strong opinionated design guidance and an excellent mandatory verification loop. Its principal defects are that both of its delegated guide files (editing.md, pptxgenjs.md) are missing from the bundle, leaving the edit/create workflows with only high-level steps, and the edit workflow omits the bundled validate.py as a pre-pack checkpoint. Moderate redundancy between the design-guidance sections and the 'Avoid' list also costs some token efficiency.
Suggestions
Ship editing.md and pptxgenjs.md in the bundle (or inline minimal command sequences for unpack/add_slide/clean/pack and a pptxgenjs starter script) so the two primary workflows don't dead-end at dangling links.
Add a validation checkpoint to the editing workflow — e.g. '3. Validate: python scripts/office/validate.py unpacked/' before the clean/pack steps — since the validator already ships in scripts/office/ and XML manipulation without validation risks a corrupt output.
De-duplicate the 'Avoid (Common Mistakes)' list against the 'Before Starting'/'For Each Slide' sections, and consider moving the palette/typography tables into a design reference file to slim the always-loaded overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and opinionated — palette tables, typography pairings, a copy-paste QA prompt, exact commands — with essentially zero space spent teaching concepts Claude already knows (no 'what is a pptx' or library tutorials). It falls short of level 5 because of redundancy: the 'Avoid (Common Mistakes)' list restates points already made in 'Before Starting'/'For Each Slide' (layout variety, color contrast, size contrast, text-only slides), and design-idea content like the 10-palette table is a reference that competes with the overview role of SKILL.md. | 4 / 5 |
Actionability | Reading, conversion, and QA are fully executable ('python -m markitdown presentation.pptx', 'python scripts/thumbnail.py presentation.pptx', 'pdftoppm -jpeg -r 150 -f N -l N output.pdf slide-fixed', a ready-to-use grep and subagent prompt). However the two core workflows delegate entirely to 'Read [editing.md](editing.md)' and 'Read [pptxgenjs.md](pptxgenjs.md)', and neither file exists in the bundle — the editing section is left with arrow shorthand ('Unpack → manipulate slides → edit content → clean → pack') and no commands for the bundled add_slide.py/clean.py. That is concrete guidance with key details missing, matching the level-3 anchor rather than level 4's 'minor gaps.' | 3 / 5 |
Workflow Clarity | The QA section is exemplary — an explicit verification loop ('Generate slides → Convert to images → Inspect → Fix issues → Re-verify affected slides → Repeat') with a hard gate ('Do not declare success until you've completed at least one fix-and-verify cycle') and placeholder-text checks — matching the level-5 pattern for feedback loops. This keeps it above level 3, but it does not reach level 5 because the editing workflow itself ('Unpack → manipulate slides → edit content → clean → pack') has no pre-pack validation checkpoint, even though scripts/office/validate.py ships in the bundle and is never mentioned. | 4 / 5 |
Progressive Disclosure | The in-document structure is good — a Quick Reference routing table, clearly signaled one-level-deep pointers ('Read editing.md', 'Read pptxgenjs.md'), and correct relative paths to real scripts (thumbnail.py, office/unpack.py, office/soffice.py). But judged against the actual bundle, both markdown references are dangling: editing.md and pptxgenjs.md do not exist anywhere in the skill directory, so navigation dead-ends for the two primary workflows; meanwhile ~90 lines of design-ideas guidance that could live in a reference file are inlined. This sits between level 2 (broken/buried references, inlined separable content) and level 4 (clean reference split), and is not clearly below the midpoint. | 3 / 5 |
Total | 14 / 20 Passed |