office-docx - SKILL.md Agent Skill

name: office-docx description: "Use when the user asks to create, edit, inspect, polish, verify, or deliver Word .docx documents, Google Docs-targeted drafts, business briefs, forms, reports, tables, checklists, redraft-ready document sections, or PDF/Word source-to-DOCX transformations." requires: bins: [python3] install:

kind: brew formula: python bins: [python3] label: "Install Python 3 via Homebrew" os: [darwin] allowed-tools:
read
write
edit
apply_patch
exec
ls
grep
pdf
web_search
web_fetch
send_attachment

Office DOCX

Use the bundled scripts in this skill package to produce editable .docx files. The builder writes deterministic OOXML with real Word numbering, tables, images with alt text, comments, and tracked changes. The skill activation metadata includes Skill directory; treat that as SKILL_DIR and run scripts from SKILL_DIR/scripts/.

Workflow

Decide the document archetype: brief, proposal, SOP, form, checklist, report, memo, or reference guide.
Create a JSON spec in the working directory. Use semantic blocks: heading, paragraph, bullet_list, numbered_list, table, callout, image, comment, revision, and page_break.
For a new document, run:

python3 "$SKILL_DIR/scripts/check_env.py"
python3 "$SKILL_DIR/scripts/build_docx.py" --spec spec.json --out output.docx
python3 "$SKILL_DIR/scripts/inspect_docx.py" --verify output.docx
python3 "$SKILL_DIR/scripts/a11y_audit.py" output.docx

To append to an existing document, use a spec containing only the new blocks, then run:

python3 "$SKILL_DIR/scripts/append_docx.py" --input existing.docx --spec append.json --out output.docx
python3 "$SKILL_DIR/scripts/inspect_docx.py" --verify output.docx

To convert a Markdown draft directly:

python3 "$SKILL_DIR/scripts/markdown_to_docx.py" --markdown draft.md --out output.docx
python3 "$SKILL_DIR/scripts/inspect_docx.py" --verify output.docx

For Google Docs-targeted output, sanitize before render/import:

python3 "$SKILL_DIR/scripts/google_docs_title_sanitize.py" output.docx --out sanitized.docx
python3 "$SKILL_DIR/scripts/google_docs_title_sanitize.py" sanitized.docx --check

Use sanitized.docx for preview rendering and native Google Docs import.

For redline/comment work, use comment blocks or revision blocks:

{"type": "comment", "target": "This sentence needs support.", "comment": "Add source."}
{"type": "revision", "delete": "old wording", "insert": "new wording"}

For delivery cleanup, use the focused helpers:

python3 "$SKILL_DIR/scripts/comments_strip.py" commented.docx --out no-comments.docx
python3 "$SKILL_DIR/scripts/comments_extract.py" commented.docx
python3 "$SKILL_DIR/scripts/add_tracked_replacements.py" --input input.docx --spec replacements.json --out redlined.docx
python3 "$SKILL_DIR/scripts/accept_tracked_changes.py" redlined.docx --mode accept --out accepted.docx
python3 "$SKILL_DIR/scripts/redact_docx.py" input.docx redacted.docx --emails --phones
python3 "$SKILL_DIR/scripts/privacy_scrub_metadata.py" input.docx --out scrubbed.docx
python3 "$SKILL_DIR/scripts/compare_docx.py" before.docx after.docx

For advanced Word structure tasks, use the focused OOXML helpers:

python3 "$SKILL_DIR/scripts/insert_toc.py" input.docx --out with-toc.docx
python3 "$SKILL_DIR/scripts/fields_report.py" with-toc.docx
python3 "$SKILL_DIR/scripts/insert_note.py" input.docx --kind footnote --text "Source note" --out with-note.docx
python3 "$SKILL_DIR/scripts/watermark_add.py" input.docx --text "DRAFT" --out watermarked.docx
python3 "$SKILL_DIR/scripts/watermark_audit_remove.py" watermarked.docx
python3 "$SKILL_DIR/scripts/set_protection.py" input.docx --mode readOnly --out protected.docx
python3 "$SKILL_DIR/scripts/content_controls.py" input.docx --spec controls.json --out form.docx
python3 "$SKILL_DIR/scripts/internal_nav.py" input.docx --spec nav.json --out linked.docx
python3 "$SKILL_DIR/scripts/merge_docx_append.py" --input a.docx --input b.docx --out merged.docx
python3 "$SKILL_DIR/scripts/docx_table_to_csv.py" input.docx --out-dir tables

If visual QA matters and LibreOffice is available, render previews:

python3 "$SKILL_DIR/scripts/render_preview.py" output.docx

Deliver the .docx path or attach it with send_attachment.

Spec Shape

{
  "title": "Document title",
  "subtitle": "Optional subtitle",
  "blocks": [
    {"type": "heading", "level": 1, "text": "Section"},
    {"type": "paragraph", "text": "Body copy."},
    {"type": "bullet_list", "items": ["Point one", {"text": "Point two", "comment": "Verify"}]},
    {"type": "image", "path": "chart.png", "alt": "Revenue trend chart", "caption": "Figure 1. Revenue trend", "width_inches": 5.5, "height_inches": 3.2},
    {"type": "revision", "delete": "Old sentence.", "insert": "Improved sentence."},
    {"type": "table", "headers": ["Metric", "Value"], "rows": [["ARR", "$1.2M"]]}
  ]
}

Quality Bar

Keep documents editable: use semantic headings, real Word lists, comments, tracked changes, images with alt text, and real tables.
Run a11y_audit.py before delivery; fix hard issues such as fake bullets, missing image alt text, empty documents, and structural defects.
Do not overuse tables for normal prose.
Avoid dense walls of text unless the document type demands it.
For Google Docs-targeted output, keep the title simple and native-looking and run google_docs_title_sanitize.py --check.
When editing existing DOCX packages, preserve package structure: append body content before the final w:sectPr, keep existing headers/relationships, and let watermark helpers allocate a new header part instead of replacing one.
If preview rendering fails because LibreOffice or a PDF-to-PNG renderer is missing, state exactly which verification passed; do not imply visual QA passed.