docx

Tier-2 Review: docx (Document Generation Cluster)

Score: 4.8 / 5.0 — A precision instrument for a narrow job, but not a general-purpose office suite.

---

The docx skill is the most tightly-scoped tool in the document-generation cluster. Compared to xlsx and docx in the same Anthropic catalog, this skill demonstrates a clear philosophy: do one thing (create and manipulate .docx files) with surgical precision, and delegate everything else to system tools. This is both its greatest strength and its most significant limitation.

Trigger Clarity (5.0) — Flawless

The trigger section is a masterclass in boundary-setting. It explicitly lists what triggers the skill ("Word doc", .docx, "report", "memo", "letter") and, crucially, what does not ("Do NOT use for PDFs, spreadsheets, Google Docs"). This is the kind of unambiguous guidance that prevents the model from drifting into territory better served by other skills. Compare this to xlsx in the same cluster, which has a more open-ended trigger list that occasionally overlaps with CSV-handling tools.

Output Specificity (5.0) — Concrete and Executable

The skill doesn't just say "create a Word document" — it provides exact JavaScript code snippets with DXA measurements for page sizes, validation commands, and conversion pipelines. The table of common page sizes (US Letter vs. A4) with precise DXA values is a small detail that saves enormous debugging time. The note that "docx-js defaults to A4, not US Letter" is exactly the kind of gotcha documentation that prevents a full document rebuild.

Scope Precision (5.0) — Honest About Its Boundaries

This skill knows what it is and what it isn't. It explicitly calls out that LibreOffice is required for accept_changes.py and PDF conversion, and that pandoc is used for text extraction. It doesn't pretend to be a full office suite — it's a document manipulation toolkit with clear dependencies. The decision to use JavaScript (docx-js) for creation and Python for validation/conversion is pragmatic, not dogmatic.

Self-Containment (5.0) — Complete, With One Caveat

The skill includes all necessary scripts (unpack.py, validate.py, soffice.py, accept_changes.py) and installation instructions. However, the harness observation reveals a critical gap: LibreOffice is not installed by default in the container. This means accept_changes.py and soffice.py will fail silently unless the environment is prepared. The skill documents this dependency, but doesn't include a setup script or fallback. This is a minor self-containment issue — the documentation is complete, but the execution environment isn't.

Reusability (4.0) — The Weak Link

This is where the skill loses a full point. The docx skill is highly reusable within its domain — any task involving .docx files benefits from its patterns. But compared to xlsx in the same catalog, which can be adapted for data analysis, reporting, and visualization, the docx skill is a one-trick pony. It creates, reads, edits, and converts .docx files, and that's it. There's no provision for template management, no integration with data sources, no batch processing patterns. The skill is a hammer, not a Swiss Army knife.

Harness Observations

From the test harness (D-048), two of three tests pass cleanly:

• install: npm install -g docx works without system libs
• minimal-roundtrip: Creates, validates, and extracts text from a 2KB document

The edge-corner test is partial: pandoc extraction works, but accept_changes.py fails if LibreOffice isn't installed. This is a real-world blocker — any task involving tracked changes acceptance will fail in a clean container. The skill should either include a LibreOffice installation step or provide a fallback method.

Verdict

The docx skill is a 4.8 because it does what it sets out to do with exceptional clarity and precision. It's the kind of skill you reach for when you need a .docx file and you need it right. But it's not a skill you reach for when you need to think about documents holistically — for that, you'd combine it with xlsx for data tables, pandoc for format conversion, and perhaps a templating engine for reuse. The skill is a brilliant soloist, but it needs an orchestra to make a symphony.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-06-04.

Overall: ok. 2 tests passed, 1 partial; key blocker: LibreOffice not installed by default, causing accept_changes.py to fail.

Inferred dependencies: node, npm, python3, pandoc, libreoffice (for accept_changes.py and soffice.py).

Test	Status	Notes
install	pass	docx npm package installs successfully; no system libs required for creation. Python scripts for validation and conversion require python3 and libreoffice, but those are separate.
minimal-roundtrip	pass	Minimal document created, validated, and text extracted. No warnings. File size ~2KB.
edge-corner	partial	pandoc extraction works; accept_changes.py requires LibreOffice (soffice) which may not be installed in container. If missing, accept step fails.