semos-agentura-document 0.5.0__tar.gz

semos_agentura_document-0.5.0/.env.example

@@ -0,0 +1,47 @@
+# --- OCR Provider: Mistral API (default) ---
+MISTRAL_API_KEY=
+
+# --- Document AI (Azure AI Foundry, auto-detected if both set) ---
+DOCUMENT_AI_ENDPOINT=
+DOCUMENT_AI_API_KEY=
+DOCUMENT_AI_MODEL=mistral-document-ai-2512
+
+# --- Diagram code generation LLM ---
+# DIAGRAM_CODEGEN_ENDPOINT=
+# DIAGRAM_CODEGEN_API_KEY=
+# DIAGRAM_CODEGEN_MODEL=
+
+# --- Diagram review LLM (falls back to document_ai if unset) ---
+# DIAGRAM_REVIEW_ENDPOINT=
+# DIAGRAM_REVIEW_API_KEY=
+# DIAGRAM_REVIEW_MODEL=
+
+# --- OCR settings ---
+# TABLE_FORMAT=markdown  # markdown or html
+# MAX_PDF_PAGES=10
+
+# --- External tool paths (optional, auto-detected on PATH) ---
+# LIBRE_OFFICE_PATH=
+# MARP_PATH=
+# PANDOC_PATH=
+# MMDC_PATH=
+# DRAWIO_PATH=
+
+# --- draw.io desktop app (for rendering diagrams with embedded images) ---
+# The npm drawio CLI cannot render inline base64 images. Set this to the
+# draw.io desktop executable for correct image rendering.
+# Windows installed:  C:\Program Files\draw.io\draw.io.exe
+# Windows portable:   C:\PortableApps\DrawioPortable\App\Drawio\draw.io.exe
+# Linux:              /usr/bin/drawio
+# macOS:              /Applications/draw.io.app/Contents/MacOS/draw.io
+# DRAWIO_DESKTOP_PATH=
+
+# --- Slide template application ---
+# TEMPLATE_BACKEND=auto  # auto, com, uno, docker
+# COM (PowerPoint) is the only backend with full corporate branding.
+# UNO/Docker transfer placeholder structure only (no logos/bars).
+
+# --- A2A LLM router (optional, enables NL delegation) ---
+# ROUTER_LLM_MODEL=
+# ROUTER_LLM_API_KEY=
+# ROUTER_LLM_API_BASE=

semos_agentura_document-0.5.0/.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+*.py[oc]
+.venv/
+.env
+tools/node_modules/
+test_forms/

semos_agentura_document-0.5.0/.python-version

@@ -0,0 +1 @@
+3.13

semos_agentura_document-0.5.0/CONTRIBUTING.md

@@ -0,0 +1,19 @@
+# Contributing to semos-agentura-document
+
+Part of the [Semos Agentura](../../README.md) monorepo. Read the
+[root CONTRIBUTING guide](../../CONTRIBUTING.md) for setup, the dev loop, and conventions.
+
+- **Import:** `from semos.agentura.document import ...`
+- **What it is:** the document agent: OCR digest, compose to PDF/PPTX/DOCX/HTML, diagram
+  generation (Mermaid, draw.io), and PDF/DOCX form inspection and filling. Depends on
+  `semos-agentura-core`.
+
+```bash
+make install                         # from the repo root, once
+cd packages/semos-agentura-document
+uv run pytest tests/                 # this package's tests
+```
+
+Tests needing external renderers (pandoc, mermaid-cli, marp, draw.io) skip when the tool is
+absent, so they pass without a full toolchain. Cross-package end-to-end tests live in the
+repo-root `tests/integration/`.

semos_agentura_document-0.5.0/Dockerfile

@@ -0,0 +1,33 @@
+# Document Agent - OCR, compose, diagrams, forms
+FROM semos-agentura-base:latest
+
+# System tools for document processing
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    pandoc \
+    texlive-xetex \
+    texlive-fonts-recommended \
+    libreoffice-calc \
+    libreoffice-writer \
+    libreoffice-impress \
+    && rm -rf /var/lib/apt/lists/*
+
+# Node.js tools (marp, mermaid-cli, draw.io-export)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    nodejs \
+    npm \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY packages/semos-agentura-document/tools/package.json /app/packages/semos-agentura-document/tools/
+RUN cd /app/packages/semos-agentura-document/tools && npm install --omit=dev \
+    && npx playwright install --with-deps chromium
+
+WORKDIR /app/packages/semos-agentura-document
+
+# Chromium needs --no-sandbox when running as root in Docker
+RUN echo '{"args":["--no-sandbox","--disable-setuid-sandbox"]}' > /app/puppeteer.json
+ENV PUPPETEER_CONFIG=/app/puppeteer.json
+
+EXPOSE 8002
+
+CMD ["uv", "run", "uvicorn", "semos.agentura.document.service:app", \
+     "--host", "0.0.0.0", "--port", "8002"]

semos_agentura_document-0.5.0/PKG-INFO

@@ -0,0 +1,295 @@
+Metadata-Version: 2.4
+Name: semos-agentura-document
+Version: 0.5.0
+Summary: Document digestion (OCR to Markdown) and composition (Markdown to documents) agent
+Requires-Python: >=3.11
+Requires-Dist: httpx>=0.28
+Requires-Dist: lxml>=5.0
+Requires-Dist: mistralai>=1.12
+Requires-Dist: pillow>=10.0
+Requires-Dist: pydantic-settings>=2.13
+Requires-Dist: pydantic>=2.0
+Requires-Dist: pypdf[crypto]>=5.0
+Requires-Dist: pypdfium2>=4.0
+Requires-Dist: python-dotenv>=1.2
+Requires-Dist: python-pptx>=1.0
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: semos-agentura-core
+Provides-Extra: com
+Requires-Dist: pywin32>=311; extra == 'com'
+Provides-Extra: image
+Requires-Dist: rembg>=2.0; extra == 'image'
+Provides-Extra: slides
+Requires-Dist: python-pptx>=1.0; extra == 'slides'
+Requires-Dist: pyyaml>=6.0; extra == 'slides'
+Description-Content-Type: text/markdown
+
+# document-agent
+
+Document digestion (OCR to Markdown), composition (Markdown to documents), diagram generation, raster image generation, and form filling. Full round-trip support for footnotes, tracked changes, comments, and styles.
+
+## Digestion
+
+Convert PDF, images, and Office documents to Markdown.
+
+- **PDF/images**: OCR via Mistral Document AI (direct or Azure AI Foundry)
+- **DOCX/ODT**: pandoc-based extraction preserving footnotes, tracked changes, comments, and document styles
+
+```bash
+# Basic digest (auto-selects pandoc for DOCX, OCR for PDF/images)
+document-agent digest document.docx
+
+# Show all tracked changes and comments
+document-agent digest document.docx --track-changes all
+
+# Force OCR pipeline for a scanned DOCX
+document-agent digest scanned.docx --mode ocr
+
+# Inline mode (base64-embedded images, prints to stdout)
+document-agent digest document.pdf --inline
+
+# With structured annotation extraction
+document-agent digest document.pdf --schema schema.py --prompt "Extract all line items"
+```
+
+### Round-trip metadata
+
+DOCX digestion preserves four types of metadata that round-trip through compose:
+
+- **Footnotes**: `[^1]` with `[^1]: definition text` at end
+- **Tracked changes**: `[text]{.insertion author="Name" date="2026-01-01"}` / `{.deletion ...}`
+- **Comments**: `[comment text]{.comment-start id="1" author="Name"}...{.comment-end id="1"}`
+- **Document styles**: YAML front matter block (fonts, sizes, colors, spacing, margins)
+
+Supported input formats: PDF, PNG, JPG, JPEG, WEBP, TIFF, BMP, DOCX, PPTX, XLSX, ODT.
+
+## Composition
+
+Convert Markdown to various output formats with style control.
+
+```bash
+# Basic compose
+document-agent compose input.md output.docx --format docx
+
+# With a reference document for style inheritance
+document-agent compose input.md output.docx --format docx \
+  --reference-doc template.docx
+
+# With headers/footers from a template (combined with YAML styles)
+document-agent compose input.md output.docx --format docx \
+  --header-footer-doc template.docx
+
+# Slides - polished (Marp, limited editability)
+document-agent compose slides.md out.pptx --format pptx --slides
+
+# Slides - draft editable (pandoc, fully editable, rough layout)
+document-agent compose slides.md out.pptx --format pptx --slides --draft
+
+# Slides - draft + corporate template (requires PowerPoint)
+document-agent compose slides.md out.pptx --format pptx --slides --draft \
+  --template corporate.pptx
+```
+
+### YAML front matter styles
+
+Define document formatting directly in the Markdown source. Styles are auto-extracted during digest and auto-applied during compose:
+
+```yaml
+---
+title: "Document Title"
+subtitle: "Optional *formatted* subtitle"
+styles:
+  page:
+    size: "A4"                    # or "Letter"
+    margin-top: "1.5cm"
+    margin-bottom: "1.5cm"
+    margin-left: "1.5cm"
+    margin-right: "1.5cm"
+  body:
+    font: "Calibri"
+    size: 11
+    spacing-before: "0.0cm"
+    spacing-after: "0.1cm"
+    line-spacing: 1.1
+  heading1:
+    font: "Calibri"
+    size: 13
+    bold: true
+    color: "000080"               # navy
+    spacing-before: "0.3cm"
+    spacing-after: "0.1cm"
+  heading2:
+    font: "Calibri"
+    size: 11
+    bold: true
+    color: "000080"
+  heading3:
+    font: "Calibri"
+    size: 11
+    bold: true
+    color: "333333"
+  table:
+    size: 9                       # also used for footnotes and captions
+    border-color: "999999"
+    border-size: 4                # eighths of a point
+    fixed: false                  # true to keep equal-width columns
+---
+```
+
+Style priority: YAML front matter > `--reference-doc` > pandoc defaults.
+
+Mermaid and draw.io diagrams in fenced code blocks are automatically rendered as images.
+
+## Diagrams
+
+Generate and modify diagrams (Mermaid or draw.io) using LLM-powered optimization.
+
+```bash
+# Generate from description
+document-agent diagram "flowchart of CI/CD pipeline" -o diagram.png
+
+# Modify existing draw.io diagram
+document-agent diagram "Change WP3 label to Digital Infrastructure" \
+  --source diagram.drawio.png -o updated.drawio.png \
+  --code-output updated.drawio
+
+# Mermaid diagram
+document-agent diagram "sequence diagram for auth flow" --type mermaid -o auth.png
+```
+
+draw.io diagrams with embedded raster images are handled automatically:
+images are stripped for LLM editing (reducing context from ~400 KB to ~8 KB)
+and restored in the output. For correct PNG rendering of embedded images,
+set `DRAWIO_DESKTOP_PATH` in `.env` (the npm CLI can't render inline images).
+
+## Image Generation
+
+Generate, edit, and extract elements from raster images using text-to-image models.
+
+### Modes
+
+- **generate**: Text-to-image from a prompt, with optional style prefix
+- **edit**: Modify an existing image with a text prompt (inpainting with optional mask for OpenAI, img2img for Flux)
+- **cut**: Extract a specific element from an image (VLM-guided bounding box + background removal, falls back to image model isolation)
+
+### Supported providers
+
+| Provider | Endpoint format | Generate | Edit | Models |
+|----------|----------------|----------|------|--------|
+| Azure OpenAI | `https://{resource}.cognitiveservices.azure.com/openai/deployments/{model}` | Yes | Yes (mask) | gpt-image-2 |
+| OpenAI | `https://api.openai.com` | Yes | Yes (mask) | gpt-image-2, dall-e-3 |
+| Azure AI Foundry | `https://{resource}.services.ai.azure.com/providers/{provider}` | Yes | Yes (img2img) | flux-2-pro, flux-2-dev |
+| Direct provider | `https://api.bfl.ai` | Yes | Yes (img2img) | flux-2-pro, flux-2-dev |
+
+Configure via environment variables:
+
+```bash
+IMAGE_GEN_ENDPOINT=https://example.cognitiveservices.azure.com/openai/deployments/gpt-image-2
+IMAGE_GEN_API_KEY=your-key
+IMAGE_GEN_MODEL=gpt-image-2
+```
+
+### Crossover: embedding images in diagrams
+
+Generated images (icons, symbols) can be embedded into draw.io diagrams via the `embeds` parameter on `generate_diagram`. Refer to each embed by its filename in the diagram description so the LLM knows where to place them.
+
+The embed mechanism reuses the existing draw.io image strip/restore pipeline: images are represented as `__IMG_N__` placeholders during LLM codegen and replaced with base64 data URIs before rendering.
+
+## Forms
+
+Inspect and fill form fields in PDF and DOCX files.
+
+```bash
+# Inspect form fields
+document-agent inspect form.pdf
+document-agent inspect form.docx --json
+
+# Fill form fields
+document-agent fill form.pdf filled.pdf --data '{"name": "John", "date": "2026-01-01"}'
+document-agent fill form.docx filled.docx --data fields.json
+```
+
+## Slide Merge
+
+Cherry-pick and merge slides from multiple PPTX sources.
+
+```bash
+# Merge specific slides by index
+document-agent merge-slides deck1.pptx:0-5 deck2.pptx:3,7,12 -o merged.pptx
+
+# Use a different base template for theme/master
+document-agent merge-slides deck1.pptx:0-5 deck2.pptx:3,7 \
+  --base template.pptx -o merged.pptx
+
+# Force python-pptx backend (portable, no PowerPoint needed)
+document-agent merge-slides deck.pptx:0-10 -o subset.pptx --backend pptx
+```
+
+Backends: COM (PowerPoint, preserves animations/transitions/media) is preferred.
+Falls back to python-pptx (portable, pure Python) when PowerPoint is unavailable.
+Install optional dependencies: `pip install document-agent[slides]`
+
+## Setup
+
+```bash
+uv sync
+cp .env.example .env
+# Edit .env with your API keys and tool paths
+```
+
+### Node.js tools (Marp, Mermaid, draw.io CLI)
+
+Marp CLI, Mermaid CLI, and draw.io export are installed locally via npm:
+
+```bash
+cd tools && npm install
+```
+
+The agent automatically discovers them in `tools/node_modules/.bin/`.
+
+### Other external tools
+
+- [Pandoc](https://pandoc.org/) - DOCX digest and document composition - install system-wide
+- [LibreOffice](https://www.libreoffice.org/) - Office format OCR fallback - install system-wide
+- [draw.io desktop](https://github.com/jgraph/drawio-desktop) - needed for rendering diagrams with embedded images. Set `DRAWIO_DESKTOP_PATH` in `.env`.
+
+## Python API
+
+```python
+from document_agent import (
+    digest, compose, compose_editable_slides,
+    merge_slides, parse_source_args,
+    OutputFormat, OutputMode,
+)
+
+# Digest a DOCX with tracked changes
+result = digest("document.docx", track_changes="all")
+print(result.markdown)
+
+# Compose with YAML styles
+result = compose("styled.md", "output.docx", OutputFormat.PDF)
+
+# Compose with reference doc
+result = compose("input.md", "output.docx", OutputFormat.DOCX,
+                 reference_doc="template.docx")
+
+# Slides - polished (Marp)
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True)
+
+# Slides - draft editable (pandoc)
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True, draft=True)
+
+# Slides - draft + corporate template
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True, draft=True,
+                 template="corporate.pptx")
+
+# Merge slides from CLI-style args
+config = parse_source_args(
+    ["deck1.pptx:0-5", "deck2.pptx:3,7"],
+    output="merged.pptx",
+)
+result = merge_slides(config, "merged.pptx")
+```

semos_agentura_document-0.5.0/README.md

@@ -0,0 +1,269 @@
+# document-agent
+
+Document digestion (OCR to Markdown), composition (Markdown to documents), diagram generation, raster image generation, and form filling. Full round-trip support for footnotes, tracked changes, comments, and styles.
+
+## Digestion
+
+Convert PDF, images, and Office documents to Markdown.
+
+- **PDF/images**: OCR via Mistral Document AI (direct or Azure AI Foundry)
+- **DOCX/ODT**: pandoc-based extraction preserving footnotes, tracked changes, comments, and document styles
+
+```bash
+# Basic digest (auto-selects pandoc for DOCX, OCR for PDF/images)
+document-agent digest document.docx
+
+# Show all tracked changes and comments
+document-agent digest document.docx --track-changes all
+
+# Force OCR pipeline for a scanned DOCX
+document-agent digest scanned.docx --mode ocr
+
+# Inline mode (base64-embedded images, prints to stdout)
+document-agent digest document.pdf --inline
+
+# With structured annotation extraction
+document-agent digest document.pdf --schema schema.py --prompt "Extract all line items"
+```
+
+### Round-trip metadata
+
+DOCX digestion preserves four types of metadata that round-trip through compose:
+
+- **Footnotes**: `[^1]` with `[^1]: definition text` at end
+- **Tracked changes**: `[text]{.insertion author="Name" date="2026-01-01"}` / `{.deletion ...}`
+- **Comments**: `[comment text]{.comment-start id="1" author="Name"}...{.comment-end id="1"}`
+- **Document styles**: YAML front matter block (fonts, sizes, colors, spacing, margins)
+
+Supported input formats: PDF, PNG, JPG, JPEG, WEBP, TIFF, BMP, DOCX, PPTX, XLSX, ODT.
+
+## Composition
+
+Convert Markdown to various output formats with style control.
+
+```bash
+# Basic compose
+document-agent compose input.md output.docx --format docx
+
+# With a reference document for style inheritance
+document-agent compose input.md output.docx --format docx \
+  --reference-doc template.docx
+
+# With headers/footers from a template (combined with YAML styles)
+document-agent compose input.md output.docx --format docx \
+  --header-footer-doc template.docx
+
+# Slides - polished (Marp, limited editability)
+document-agent compose slides.md out.pptx --format pptx --slides
+
+# Slides - draft editable (pandoc, fully editable, rough layout)
+document-agent compose slides.md out.pptx --format pptx --slides --draft
+
+# Slides - draft + corporate template (requires PowerPoint)
+document-agent compose slides.md out.pptx --format pptx --slides --draft \
+  --template corporate.pptx
+```
+
+### YAML front matter styles
+
+Define document formatting directly in the Markdown source. Styles are auto-extracted during digest and auto-applied during compose:
+
+```yaml
+---
+title: "Document Title"
+subtitle: "Optional *formatted* subtitle"
+styles:
+  page:
+    size: "A4"                    # or "Letter"
+    margin-top: "1.5cm"
+    margin-bottom: "1.5cm"
+    margin-left: "1.5cm"
+    margin-right: "1.5cm"
+  body:
+    font: "Calibri"
+    size: 11
+    spacing-before: "0.0cm"
+    spacing-after: "0.1cm"
+    line-spacing: 1.1
+  heading1:
+    font: "Calibri"
+    size: 13
+    bold: true
+    color: "000080"               # navy
+    spacing-before: "0.3cm"
+    spacing-after: "0.1cm"
+  heading2:
+    font: "Calibri"
+    size: 11
+    bold: true
+    color: "000080"
+  heading3:
+    font: "Calibri"
+    size: 11
+    bold: true
+    color: "333333"
+  table:
+    size: 9                       # also used for footnotes and captions
+    border-color: "999999"
+    border-size: 4                # eighths of a point
+    fixed: false                  # true to keep equal-width columns
+---
+```
+
+Style priority: YAML front matter > `--reference-doc` > pandoc defaults.
+
+Mermaid and draw.io diagrams in fenced code blocks are automatically rendered as images.
+
+## Diagrams
+
+Generate and modify diagrams (Mermaid or draw.io) using LLM-powered optimization.
+
+```bash
+# Generate from description
+document-agent diagram "flowchart of CI/CD pipeline" -o diagram.png
+
+# Modify existing draw.io diagram
+document-agent diagram "Change WP3 label to Digital Infrastructure" \
+  --source diagram.drawio.png -o updated.drawio.png \
+  --code-output updated.drawio
+
+# Mermaid diagram
+document-agent diagram "sequence diagram for auth flow" --type mermaid -o auth.png
+```
+
+draw.io diagrams with embedded raster images are handled automatically:
+images are stripped for LLM editing (reducing context from ~400 KB to ~8 KB)
+and restored in the output. For correct PNG rendering of embedded images,
+set `DRAWIO_DESKTOP_PATH` in `.env` (the npm CLI can't render inline images).
+
+## Image Generation
+
+Generate, edit, and extract elements from raster images using text-to-image models.
+
+### Modes
+
+- **generate**: Text-to-image from a prompt, with optional style prefix
+- **edit**: Modify an existing image with a text prompt (inpainting with optional mask for OpenAI, img2img for Flux)
+- **cut**: Extract a specific element from an image (VLM-guided bounding box + background removal, falls back to image model isolation)
+
+### Supported providers
+
+| Provider | Endpoint format | Generate | Edit | Models |
+|----------|----------------|----------|------|--------|
+| Azure OpenAI | `https://{resource}.cognitiveservices.azure.com/openai/deployments/{model}` | Yes | Yes (mask) | gpt-image-2 |
+| OpenAI | `https://api.openai.com` | Yes | Yes (mask) | gpt-image-2, dall-e-3 |
+| Azure AI Foundry | `https://{resource}.services.ai.azure.com/providers/{provider}` | Yes | Yes (img2img) | flux-2-pro, flux-2-dev |
+| Direct provider | `https://api.bfl.ai` | Yes | Yes (img2img) | flux-2-pro, flux-2-dev |
+
+Configure via environment variables:
+
+```bash
+IMAGE_GEN_ENDPOINT=https://example.cognitiveservices.azure.com/openai/deployments/gpt-image-2
+IMAGE_GEN_API_KEY=your-key
+IMAGE_GEN_MODEL=gpt-image-2
+```
+
+### Crossover: embedding images in diagrams
+
+Generated images (icons, symbols) can be embedded into draw.io diagrams via the `embeds` parameter on `generate_diagram`. Refer to each embed by its filename in the diagram description so the LLM knows where to place them.
+
+The embed mechanism reuses the existing draw.io image strip/restore pipeline: images are represented as `__IMG_N__` placeholders during LLM codegen and replaced with base64 data URIs before rendering.
+
+## Forms
+
+Inspect and fill form fields in PDF and DOCX files.
+
+```bash
+# Inspect form fields
+document-agent inspect form.pdf
+document-agent inspect form.docx --json
+
+# Fill form fields
+document-agent fill form.pdf filled.pdf --data '{"name": "John", "date": "2026-01-01"}'
+document-agent fill form.docx filled.docx --data fields.json
+```
+
+## Slide Merge
+
+Cherry-pick and merge slides from multiple PPTX sources.
+
+```bash
+# Merge specific slides by index
+document-agent merge-slides deck1.pptx:0-5 deck2.pptx:3,7,12 -o merged.pptx
+
+# Use a different base template for theme/master
+document-agent merge-slides deck1.pptx:0-5 deck2.pptx:3,7 \
+  --base template.pptx -o merged.pptx
+
+# Force python-pptx backend (portable, no PowerPoint needed)
+document-agent merge-slides deck.pptx:0-10 -o subset.pptx --backend pptx
+```
+
+Backends: COM (PowerPoint, preserves animations/transitions/media) is preferred.
+Falls back to python-pptx (portable, pure Python) when PowerPoint is unavailable.
+Install optional dependencies: `pip install document-agent[slides]`
+
+## Setup
+
+```bash
+uv sync
+cp .env.example .env
+# Edit .env with your API keys and tool paths
+```
+
+### Node.js tools (Marp, Mermaid, draw.io CLI)
+
+Marp CLI, Mermaid CLI, and draw.io export are installed locally via npm:
+
+```bash
+cd tools && npm install
+```
+
+The agent automatically discovers them in `tools/node_modules/.bin/`.
+
+### Other external tools
+
+- [Pandoc](https://pandoc.org/) - DOCX digest and document composition - install system-wide
+- [LibreOffice](https://www.libreoffice.org/) - Office format OCR fallback - install system-wide
+- [draw.io desktop](https://github.com/jgraph/drawio-desktop) - needed for rendering diagrams with embedded images. Set `DRAWIO_DESKTOP_PATH` in `.env`.
+
+## Python API
+
+```python
+from document_agent import (
+    digest, compose, compose_editable_slides,
+    merge_slides, parse_source_args,
+    OutputFormat, OutputMode,
+)
+
+# Digest a DOCX with tracked changes
+result = digest("document.docx", track_changes="all")
+print(result.markdown)
+
+# Compose with YAML styles
+result = compose("styled.md", "output.docx", OutputFormat.PDF)
+
+# Compose with reference doc
+result = compose("input.md", "output.docx", OutputFormat.DOCX,
+                 reference_doc="template.docx")
+
+# Slides - polished (Marp)
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True)
+
+# Slides - draft editable (pandoc)
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True, draft=True)
+
+# Slides - draft + corporate template
+result = compose("slides.md", "out.pptx", OutputFormat.PPTX,
+                 is_slides=True, draft=True,
+                 template="corporate.pptx")
+
+# Merge slides from CLI-style args
+config = parse_source_args(
+    ["deck1.pptx:0-5", "deck2.pptx:3,7"],
+    output="merged.pptx",
+)
+result = merge_slides(config, "merged.pptx")
+```

semos_agentura_document-0.5.0/docker/Dockerfile.libreoffice

@@ -0,0 +1,10 @@
+FROM ubuntu:24.04
+ENV DEBIAN_FRONTEND=noninteractive
+RUN apt-get update && \
+    apt-get install -y \
+    libreoffice-impress libreoffice-script-provider-python \
+    python3-uno python3 && \
+    apt-get clean && rm -rf /var/lib/apt/lists/*
+ENV SAL_USE_VCLPLUGIN=svp
+ENV HOME=/tmp
+WORKDIR /work