Content
70%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 dense, highly operational body with an excellent phased workflow and explicit validation gates, but it runs hot on tokens: version-history and regulatory minutiae sit inline instead of in reference files, and the carefully designed reference map points at files that are absent from this bundle, so progressive disclosure cannot actually execute as written.
Suggestions
Move time-sensitive version/date detail (Pandoc 3.9 release notes, Typst 0.14 features, wkhtmltopdf EOL dates, ADA Title II deadlines) out of SKILL.md into the relevant reference files (e.g. conversion-matrix.md, accessibility-guide.md), keeping only the current recommendation in the body — this addresses the conciseness penalty for inline version numbers.
Include one or two assembled, copy-paste-ready example commands in the body (e.g. a complete `pandoc --from markdown --to pdf --pdf-engine=xelatex` invocation with defaults file) so the most common case is executable without loading a reference file.
Ship the referenced bundle files (reference/*.md and the _common/*.md files) alongside SKILL.md, or trim the Reference Map to files that exist — the progressive disclosure design is only as good as the files it can actually load.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body largely assumes competence — no 'what is a PDF' filler, and dense tables of operational rules — but it carries substantial detail inline that belongs in reference files, and the rubric explicitly penalizes inline time-sensitive version/date material ("Pandoc 3.9 (released February 4, 2026)", "Typst 0.14 (October 24, 2025)", wkhtmltopdf EOL dates, ADA Title II deadlines) that is not placed in an 'old patterns'/'deprecated' section. Mostly efficient but noticeably over budget at ~235 lines; not a 2 because almost none of it is explanation of concepts Claude already knows. | 3 / 5 |
Actionability | Concrete, executable-level guidance throughout: explicit tool+engine pairings ("pandoc + xelatex", "pandoc + weasyprint", "mermaid-cli"), concrete flags ("Use `--from` and `--to` flags explicitly"), quantified gates ("B: 80-89", "Test with 3-5 sample files"), and a dispatch table mapping subcommands to 'Read First' references that hold the actual commands. Not a 5 because no assembled, copy-paste-ready command example appears in the body itself; not a 3 because the selection guidance is specific and the concrete commands are deliberately and clearly delegated to named reference files. | 4 / 5 |
Workflow Clarity | The ANALYZE → CONFIGURE → CONVERT → VERIFY → DELIVER → TRANSMUTE workflow is a clear table with a required outcome per phase, and validation is explicit and looped: a VERIFY phase producing "A pass/fail decision or required fixes", a minimum passing grade (B / 80+), quality-score weights, grade gates, and a batch pre-flight rule ("Test with 3-5 sample files before batch conversion"). This matches the top anchor — explicit validation steps and feedback loops, including for batch operations. | 5 / 5 |
Progressive Disclosure | The reference structure is well designed — a Reference Map of one-level-deep files each with a 'Read this when...' signal, a Recipes table with a 'Read First' column, and 'Behavior depth lives in the Behavior column; load only the Read First files' — but scored against the actual bundle: no reference/, references/, scripts/, or assets/ files exist, so the ~20 referenced paths (e.g. reference/pandoc-recipes.md, _common/OPUS_5_AUTHORING.md) cannot be loaded or verified, and the SKILL.md itself is long enough that significant content is effectively inlined. Structure is present and well-signaled but unverified and over-inline — matching the anchor 'Some structure but could be better organized', not 4 because the bundle cannot actually deliver the promised progressive disclosure. | 3 / 5 |
Total | 15 / 20 Passed |