Agent Skills

Agent Skills

Skills are specialized capabilities that extend what AI agents can do. They’re stored in agent-harness/skills/ and provide tooling for document processing, data manipulation, and more.

Available Skills

DOCX - Word Document Processing

Purpose: Create, edit, and analyze Microsoft Word documents (.docx)

Capabilities:

• Create new documents with formatting, headers, tables
• Edit existing documents while preserving tracked changes
• Extract text content via pandoc
• Add comments and work with redlines
• Access raw XML for advanced manipulation

Key workflows:

• New document: Use docx-js (JavaScript) to create from scratch
• Edit existing: Use OOXML Python library with unpack/edit/pack cycle
• Redlining: Required for legal, academic, business, or government docs

---

PDF - PDF Manipulation

Purpose: Extract, create, merge, split, and fill PDF documents

Capabilities:

• Extract text and tables (pdfplumber)
• Create new PDFs (reportlab)
• Merge and split documents (pypdf, qpdf)
• Fill PDF forms (pdf-lib, pypdf)
• OCR scanned documents (pytesseract)
• Add watermarks and password protection

Key tools:

Task	Tool
Extract text	pdfplumber or pdftotext
Extract tables	pdfplumber → pandas
Create PDFs	reportlab
Merge/split	pypdf or qpdf
Fill forms	See forms.md in skill
OCR	pytesseract + pdf2image

---

PPTX - PowerPoint Presentations

Purpose: Create, edit, and analyze PowerPoint presentations (.pptx)

Capabilities:

• Create new presentations from HTML slides
• Edit existing presentations via XML manipulation
• Extract text content via markitdown
• Access speaker notes and layouts
• Generate thumbnail grids for validation

Key workflows:

• New presentation: Use html2pptx workflow (HTML → PPTX)
• Edit existing: Unpack OOXML, edit XML, validate, repack
• Text extraction: python -m markitdown file.pptx

Slide dimensions: 720pt × 405pt for 16:9 aspect ratio

---

XLSX - Excel Spreadsheets

Purpose: Create, edit, and analyze Excel spreadsheets (.xlsx)

Capabilities:

• Read and analyze data with pandas
• Create new spreadsheets with formulas and formatting
• Edit existing files while preserving formulas
• Recalculate formulas via LibreOffice
• Data visualization and analysis

Critical rule: Always use Excel formulas, never hardcode calculated values.

# ❌ Wrong
sheet['B10'] = df['Sales'].sum()

# ✅ Correct
sheet['B10'] = '=SUM(B2:B9)'

Financial model colors:

• Blue text: Hardcoded inputs
• Black text: Formulas
• Green text: Links from other sheets
• Yellow background: Key assumptions

---

Repo Structure

agent-harness/skills/
├── docx/
│   ├── SKILL.md          # Main documentation
│   ├── docx-js.md        # Creating new docs
│   ├── ooxml.md          # Editing existing docs
│   ├── ooxml/scripts/    # pack, unpack, validate
│   └── scripts/          # Python document library
├── pdf/
│   ├── SKILL.md          # Main documentation
│   ├── forms.md          # Form filling guide
│   ├── reference.md      # Advanced reference
│   └── scripts/          # PDF utilities
├── pptx/
│   ├── SKILL.md          # Main documentation
│   ├── html2pptx.md      # HTML to PPTX workflow
│   ├── ooxml.md          # OOXML editing
│   └── scripts/          # Conversion utilities
└── xlsx/
    ├── SKILL.md          # Main documentation
    └── recalc.py         # Formula recalculation

MCP Servers

MCP (Model Context Protocol) server configs are stored under agent-harness/mcp/. Each server has:

• README.md - Purpose, auth, enabling instructions
• config.json - Server configuration
• docs.md - Detailed documentation

Adding New Skills

1. Create folder under skills/<skill-name>/
2. Add SKILL.md with overview and workflows
3. Include any reference docs and scripts
4. Update AGENTS.md if workspace footprint changes