Sdo DOCX Operations

SDO DOCX Operations -- officecli guardrails

Use this skill for reliable, repeatable officecli operations that require XML-level edits on standards documents.

Workflow

1. Open the document in resident mode before multi-step edits: officecli open <doc>
2. Run one mutation command at a time and check exit codes.
3. Use style/ordinal or explicit XPath targeting, not paraId targeting.
4. Prefer officecli batch for template population, text updates, and paragraph additions. Python zipfile/ElementTree scripts can strip namespace declarations from the root element (see "Python OOXML manipulation — namespace caveat" below). Reserve Python scripts for edits that officecli cannot express.
5. Validate with officecli view/get/query before and after mutation.
6. Close the document at the end: officecli close <doc>

Script entry points

• scripts/add_bookmarks.py -- figure/table/reference bookmark insertion and reference run replacement.
• scripts/fix_heading_tabs.py -- heading number-tab-title run repair.
• scripts/fix_enum_tabs.py -- enumeration dash-tab-text run repair.
• scripts/fix_note_tabs.py -- NOTE number-tab-text run repair and sequence validation.
• scripts/officecli_xml_common.py -- shared helpers (run_officecli, raw_set, validate_doc, load_json_spec).

All scripts declare PEP 723 inline metadata and are run via uv run:

uv run scripts/add_bookmarks.py --doc <doc> --spec assets/specs/add_bookmarks.example.json
uv run scripts/fix_heading_tabs.py --doc <doc> --mapping assets/specs/fix_heading_tabs.example.json
uv run scripts/fix_enum_tabs.py --doc <doc> --items assets/specs/fix_enum_tabs.example.json
uv run scripts/fix_note_tabs.py --doc <doc> --notes assets/specs/fix_note_tabs.example.json

JSON specs

• assets/schemas/ -- JSON Schema contracts for each script input.
• assets/specs/ -- example specs for reuse across projects.

Idempotent updates — avoid hanging paragraphs

When a script inserts paragraphs (figures, tables) that are later refreshed by re-running the same script, hanging paragraphs accumulate — old inserted elements are left behind and new ones are added on each run.

Rules:

• Before inserting Figure-style image paragraphs: query and remove ALL existing style=Figure paragraphs (exact match, not contains).
  • Use paragraph[style=Figure] — not paragraph[style~=Figure].
  • Captions (style=Figure_No & title / FigureNotitle0) are updated in-place via set; do NOT remove them — they serve as stable anchor paraIds for the insertion --before target.
• Before rebuilding a table section: remove all dynamic tables by iterating remove /body/tbl[N] in a while True / except RuntimeError: break loop rather than a fixed count. A fixed count fails on repeated runs once tables have already been removed.
• Before removing paragraphs by paraId: wrap each removal in try/except (or check existence first) so the script is safe to run multiple times. ParaIds that were already removed on a previous run must not abort the script.
• Exact vs contains style match matters:
  • paragraph[style=Figure] — exact match (image containers only)
  • paragraph[style~=Figure_No] — contains match (would also remove captions)

Safety rules

• Never use raw-set --action replace with an empty XML payload.
• Always keep bookmark names and IDs unique.
• Always preserve non-breaking spaces and explicit tab runs where required.
• Keep scripts platform-independent and executable via uv run.
• Always use uv run python, never bare python.

Python OOXML manipulation — namespace caveat

When using Python zipfile + ElementTree to manipulate OOXML directly, ET.register_namespace() must be called for every namespace prefix used in the document — otherwise ET.tostring() strips undeclared prefixes from the root element, causing OpenXML schema validation errors and potentially breaking the document in Word.

Safer alternative: Prefer officecli batch for template population, text updates, and paragraph additions. officecli batch preserves all namespace declarations automatically and handles the OOXML schema correctly. Reserve Python zipfile/ElementTree for cases where officecli cannot express the required edit (e.g., inserting hyperlink relationships in .rels files).

# Required namespace registrations for even a simple docx:
ET.register_namespace(
    "w", "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
)
ET.register_namespace(
    "r", "http://schemas.openxmlformats.org/officeDocument/2006/relationships"
)
ET.register_namespace("w14", "http://schemas.microsoft.com/office/word/2010/wordml")
ET.register_namespace("w15", "http://schemas.microsoft.com/office/word/2012/wordml")
# ... and potentially 20+ more (wp14, w16, w16se, w16cid, w16sdtdh, etc.)

If even one namespace is missing, the document's root element loses those declarations on serialization and the file fails validation.

• Use extract_paragraphs() from references/raw-xml-manipulation.md — never the naive <w:p ...>.*?</w:p> regex (breaks on self-closing <w:p/>).
• When extracting <w:rPr>, use balanced <w:rPr> token matching — never find('<w:rPr', ...) which also matches <w:rPrChange>.
• Validate replacement XML with validate_para_xml() before batching.
• Check for ' REF REF_' in pxml before tag-scanning to avoid double-replacement after a paragraph is already fixed.

References

• references/tool-usage.md
• references/constraints.md
• references/xpath-patterns.md
• references/field-code-patterns.md
• references/tracked-changes.md
• references/raw-xml-manipulation.md

Cross-references

• sdo-docx-formatting -- formatting rules that these operations implement
• 3gpp-drafting / etsi-drafting / itut-drafting -- concrete style names required for XPath targeting (e.g. Heading1, B1, NO, TF)
• docx-svg -- SVG image insertion when officecli is not available