Content
57%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-structured, largely actionable skill whose strongest workflow (redlining) has explicit batching and end-to-end verification. Its weaknesses are duplicated batching guidance, a validation-free basic editing workflow that overwrites documents, and references plus script paths that point to files missing from this bundle.
Suggestions
Add a validation/verification step to the 'Basic OOXML editing' workflow (e.g., pandoc conversion or an XML schema check of the unpacked directory between editing and packing) to catch errors before the original .docx is overwritten.
Fix the dangling references: docx-js.md and ooxml.md are not present in the bundle, and ooxml/scripts/unpack.py / pack.py do not match the actual scripts/ directory (document.py, utilities.py, templates/) — either ship the referenced files or update the paths and remove the mandatory-read instructions.
Deduplicate the batch-grouping guidance, which currently appears three times in the redlining section ('Batching Strategy', step 2 'Batch organization', and step 4 'Suggested batch groupings'); state it once.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient (decision tree, copy-paste commands, minimal concept explanation), but the redlining section repeats batch-grouping guidance three times — the 'Batching Strategy' paragraph, step 2's 'Batch organization' list, and step 4's 'Suggested batch groupings' — and the 'MANDATORY - READ ENTIRE FILE' admonition appears twice in nearly identical ALL-CAPS form. This fits 'mostly efficient but could be tightened' rather than the 'minor instances' of anchor 4. | 3 / 5 |
Actionability | Most guidance is concrete and executable (pandoc/unpack/pack/soffice/pdftoppm commands with flags, a worked tracked-changes XML snippet, explicit word_read/word_edit call shapes). However, the two core workflows ('Creating a new Word document' and 'Editing an existing Word document') defer all actual code patterns to docx-js.md and ooxml.md, which are not present in the bundle, so a reader cannot execute those workflows from the body alone — 'concrete code or commands with minor gaps'. | 4 / 5 |
Workflow Clarity | Sequencing is strong (decision tree up front; numbered redlining steps with a final verification step of pandoc conversion plus grep assertions and an unintended-changes check; live-editing retry on revision conflict). But the rubric caps workflow clarity at 3 when a destructive/batch workflow lacks validation, and the 'Basic OOXML editing' workflow (unpack → script → pack) modifies and overwrites a document with no validation or output verification step. | 3 / 5 |
Progressive Disclosure | SKILL.md itself is well-sectioned and its references ([docx-js.md](docx-js.md), [ooxml.md](ooxml.md)) are clearly signaled and one level deep, but both referenced files are absent from the bundle (no references/ directory, no .md files beyond SKILL.md), and the invoked ooxml/scripts/unpack.py and pack.py paths do not match the actual scripts/ layout (document.py, utilities.py, templates/). The 'NEVER set any range limits' full-file-read instruction also pulls ~1100 lines into context wholesale, undermining the reference pattern's purpose. Structure is present but organization is compromised by dangling references. | 3 / 5 |
Total | 13 / 20 Passed |