| name | doc |
| description | Create, inspect, and edit Word documents (`.docx`) with local code-execution tools. Use for document drafting, structured edits, tables, comments, and layout-sensitive Word deliverables. |
| license | Apache-2.0 |
| metadata | {"author":"Open Interpreter","managed-by":"interpreter-workstation"} |
Word documents
This is workflow guidance, not a callable tool. Use the shell or code-execution tool exposed by the active OIX harness. Do not look for a tool named doc.
Workflow
- Inspect the source before editing. Preserve its structure, styles, identifiers, and branding unless the user asks for a redesign.
- Use
python-docx for creation and structured edits. For unsupported features such as tracked changes or complex comments, edit OOXML deliberately and keep the change narrowly scoped.
- Write one cohesive script for a substantial change instead of a chain of fragile one-line mutations.
- Save to a new file while iterating unless the user explicitly asked to replace the original.
- Verify content by reopening the saved document with
python-docx and checking the requested text, tables, and section structure.
- When layout matters, render every page with
scripts/render_docx.py or an available permissive office renderer, inspect the images, fix defects, and render again.
- If the file is open in Interpreter, call
interpreter-app tools builtin-interpreter interpreter_refresh_file --json '{"path":"/absolute/path.docx"}' once after the final write.
Dependencies
Prefer uv and install only what the task needs:
uv pip install python-docx
Use an existing Python installation when available. If installation is blocked, ask for approval for the exact dependency command.
Quality bar
- Use named styles and a consistent type, spacing, margin, and heading system.
- Keep tables within page bounds; avoid clipped text, broken pagination, and accidental blank pages.
- Preserve source identity and exact requested facts.
- Keep scratch files outside the final deliverable folder.
- Reopen the final
.docx and inspect all rendered pages before delivery when rendering is available.
This skill is derived from an Apache-2.0 OpenAI document workflow and has been modified for OIX code execution and Workstation refresh behavior.
