libreoffice-writer（自由办公室作家）

LibreOffice Writer

Use the bundled writer modules for UNO-backed Writer document work. All paths must be absolute. Bundled modules live under scripts/ in this skill directory, so set PYTHONPATH=<skill_base_dir>/scripts. If setup or runtime issues appear, check references/troubleshooting.md.

API Surface

# Non-session utilities
create_document(path, source=None)              # source: path to .md file to import
export_document(path, output_path, export_format)   # formats: "pdf", "docx", "md"
snapshot_page(doc_path, output_path, page=1, dpi=150)

# Session (primary editing API)
WriterSession(path) -> context manager

WriterSession methods:
  read_text(target: WriterTarget | None = None) -> str
  insert_text(text, target: WriterTarget | None = None)
  replace_text(target: WriterTarget, new_text)
  delete_text(target: WriterTarget)
  format_text(target: WriterTarget, formatting: TextFormatting)
  insert_table(rows, cols, data=None, name=None, target: WriterTarget | None = None)
  update_table(target: WriterTarget, data)
  delete_table(target: WriterTarget)
  insert_image(image_path, width=None, height=None, name=None, target: WriterTarget | None = None)
  update_image(target: WriterTarget, image_path=None, width=None, height=None)
  delete_image(target: WriterTarget)
  insert_list(items: list[ListItem], ordered: bool, target: WriterTarget | None = None)
  replace_list(target: WriterTarget, items: list[ListItem], ordered: bool | None = None)
  delete_list(target: WriterTarget)
  set_metadata(values: dict[str, str])
  get_metadata() -> dict[str, str]
  patch(patch_text, mode="atomic") -> PatchApplyResult
  export(output_path, export_format)
  reset()
  close(save=True)

# Standalone patch utility
patch(path, patch_text, mode="atomic") -> PatchApplyResult

Structured Targets: WriterTarget

from writer import WriterTarget

WriterTarget(
    kind="text" | "insertion" | "table" | "image" | "list",
    text=None,
    after=None,
    before=None,
    occurrence=None,
    name=None,
    index=None,
)

Target kinds

Resolution rules

• Omit target to read the full document or append inserted content at the end.
• Use after and before to constrain a search window.
• Use occurrence when repeated text is expected; otherwise matching must be unique.
• Prefer full sentences or distinctive paragraph-sized phrases for text, after, and before anchors; single-word anchors are often too brittle for realistic prose edits.
• For table/image targets, prefer name; use index only when order is stable.
• For insertion after inline text, Writer inserts at the boundary after the matched span; paragraph breaks must come from the inserted text or the session helper.

Formatting Payload

from writer import TextFormatting

TextFormatting(
    bold=None,
    italic=None,
    underline=None,
    font_name=None,
    font_size=None,
    color=None,          # named color or integer
    align=None,          # "left" | "center" | "right" | "justify" | "start" | "end"
    line_spacing=None,
    spacing_before=None,
    spacing_after=None,
)

Notes:

• Character and paragraph formatting can be combined in one call.
• Paragraph properties such as align apply to the full paragraph containing the match, not just the exact matched span.
• At least one formatting field must be set.

List Items

from writer import ListItem

ListItem(text="Confirm scope", level=0)

• level is zero-based nesting.
• Nesting cannot skip levels.
• ordered=True uses a numbering style; ordered=False uses bullets.

Metadata

• set_metadata() accepts a dictionary of key-value pairs.
• Supported keys are title, author, subject, keywords, and description.
• keywords may be a comma-separated string; get_metadata() returns them as one comma-separated string.

Patch DSL

Use patch() or session.patch() to apply ordered operations.

[operation]
type = format_text
target.kind = text
target.text = Quarterly revenue grew 18%.
target.after = Financial Summary
target.before = Action Items
format.bold = true
format.align = center

[operation]
type = insert_list
target.kind = insertion
target.after = Action Items
list.ordered = false
items <<JSON
[
  {"text": "Confirm scope", "level": 0},
  {"text": "Review output", "level": 0},
  {"text": "Update packaging", "level": 1}
]
JSON

Supported operation types

• insert_text
• replace_text
• delete_text
• format_text
• insert_table
• update_table
• delete_table
• insert_image
• update_image
• delete_image
• insert_list
• replace_list
• delete_list

Patch value rules

• Use target.* fields for target definition.
• To append at the end of the document, omit all target.* fields; do not use a bare target.kind = insertion without anchors.
• Use format.* fields for formatting payloads.
• Use list.ordered plus JSON items for list operations.
• items and data must be valid JSON.
• Heredoc blocks are supported with <<TAG... TAG for multiline text or JSON.

Modes

• atomic stops on first failure, resets the session, and persists nothing.
• best_effort keeps successful earlier operations and records failures.

PatchApplyResult fields:

• mode
• overall_status = "ok" | "partial" | "failed"
• operations = list of PatchOperationResult
• document_persisted

For standalone patch(path,...), document_persisted means the changes were saved to disk. For session.patch(...), it means the patch produced successful mutations in the current open session state.

Example: Edit a Report in Session

from pathlib import Path

from writer import ListItem, TextFormatting, WriterSession, WriterTarget
from writer.core import create_document

output = str(Path("test-output/report.odt").resolve())
create_document(output)

with WriterSession(output) as session:
    session.insert_text(
        "Executive Summary\n\n"
        "Financial Summary\n\n"
        "Quarterly revenue grew 18%.\n\n"
        "Action Items"
    )
    session.format_text(
        WriterTarget(
            kind="text",
            text="Quarterly revenue grew 18%.",
            after="Financial Summary",
            before="Action Items",
        ),
        TextFormatting(bold=True, align="center"),
    )
    session.insert_list(
        [
            ListItem(text="Confirm scope", level=0),
            ListItem(text="Review output", level=0),
            ListItem(text="Update packaging", level=1),
        ],
        ordered=False,
        target=WriterTarget(kind="insertion", after="Action Items"),
    )

Example: Patch an Existing Document

from writer import patch

result = patch(
    "/abs/path/report.odt",
    """
    [operation]
    type = replace_text
    target.kind = text
    target.text = Draft
    new_text = Final

    [operation]
    type = update_table
    target.kind = table
    target.name = Summary
    data = [["Metric", "Value"], ["Revenue", "$2M"]]

    [operation]
    type = replace_list
    target.kind = list
    target.text = Confirm scope
    items = [{"text": "Approve release", "level": 0}, {"text": "Notify team", "level": 1}]
    list.ordered = true
    """,
    mode="best_effort",
)

print(result.overall_status)

Snapshots

from pathlib import Path
from writer import snapshot_page

result = snapshot_page(doc_path, "/tmp/page1.png", page=1, dpi=150)
print(result.file_path, result.width, result.height)
Path(result.file_path).unlink(missing_ok=True)   # clean up used snapshots

Use snapshots to verify layout after formatting, list edits, image placement, or table changes.

• Prefer session.export() while a document is already open in a WriterSession.
• Use standalone export_document() when exporting an existing file without keeping a session open.
• get_metadata() normalizes keywords into one comma-and-space separated string.

Common Mistakes

• Passing a relative path; UNO-facing Writer APIs expect absolute file paths.
• Omitting occurrence for repeated text and then getting an ambiguity error.
• Using anchors that are too short or too common; prefer full-sentence or paragraph-level anchor text plus after / before bounds when possible.
• Expecting align to apply only to a phrase; Writer applies paragraph alignment to the containing paragraph.
• Supplying malformed JSON in items or data patch fields.
• Forgetting to clean up captured snapshots after inspection.
• Calling session.export() or other methods after session.close().
• The Markdown filter for markdown import / export requires LibreOffice 26.2+.
• The "start" and "end" paragraph alignment options require LibreOffice 26.2+.