Skills & Tools

Your agent comes with pre-installed skills that extend its capabilities. Skills are documentation and tooling that teach your agent how to use specific tools and APIs.

Pre-installed skills

Skills are installed in two locations:

fat-skills (Platform skills)

Installed at /opt/fat-skills/ from tinyfatco/fat-skills:

Skill	Description
browser-tools	Headless Chrome — screenshots, navigation, JS evaluation
gh-login	GitHub CLI authentication setup
pdf-gen	PDF generation with headless Chromium

Agent skills

Three-tier priority system:

1. System skills (/opt/fat-skills/) — Platform-provided, lowest priority
2. Workspace skills (/data/skills/) — Per-agent custom skills
3. Channel skills (/data/{channelId}/skills/) — Per-channel overrides, highest priority

User skills always override system skills of the same name.

Built-in tools

Your agent has access to core tools without any additional setup:

File operations

• Read and write files anywhere in /data
• Create directories
• List directory contents

Shell commands

• Run any Linux command via bash
• Access to common utilities (see below)

Code editing

• Surgical file edits (find and replace)
• Full file rewrites

Container utilities

The sandbox container includes these pre-installed tools:

Core:

• git — Version control
• curl, wget — HTTP requests
• jq, yq — JSON/YAML processing
• ripgrep (rg) — Fast text search
• fd-find (fdfind) — Fast file finding
• tree, less — File browsing

Databases:

• sqlite3 — SQLite client
• psql — PostgreSQL client

Python:

• python3 — Python interpreter
• uv — Fast Python package manager
• fpdf2 — PDF generation (pre-installed)

Developer:

• gh — GitHub CLI
• tmux — Terminal multiplexer
• Node.js 22 LTS

Browser:

• Headless Chromium — Screenshots, navigation, PDF generation
• browser-screenshot.js — Take screenshots
• browser-nav.js — Navigate to URLs
• browser-eval.js — Run JavaScript on pages

Browser automation

Your agent can use headless Chrome for web tasks:

# Navigate to a URL
browser-nav.js https://example.com

# Take a screenshot
browser-screenshot.js

# Run JavaScript on the page
browser-eval.js "document.title"

This enables:

• Taking screenshots of websites
• Generating PDFs from HTML
• Scraping web content
• Filling out forms

Note

Browser tools require the standard-1 container instance type (2 GiB RAM). This is the default for TinyFat containers.

Adding capabilities

Install Python packages

uv pip install pandas matplotlib

Packages persist in your agent’s storage.

Install Node packages

npm install -g some-package

Note

Installed packages persist across turns but may not survive container updates. For critical dependencies, add installation to your MEMORY.md instructions.

API keys

Some skills require API keys:

Skill	Key needed	How to add
Brave Search	BRAVE_API_KEY	Dashboard → Settings
GitHub CLI	GITHUB_TOKEN	Dashboard → Settings (or via gh-login skill)

PDF generation

Two options:

Headless Chrome (recommended)

browser-nav.js file:///data/report.html
browser-screenshot.js --pdf /data/report.pdf

Python fpdf2

from fpdf import FPDF

pdf = FPDF()
pdf.add_page()
pdf.set_font("helvetica", size=12)
pdf.cell(200, 10, text="Hello World!", new_x="LMARGIN", new_y="NEXT")
pdf.output("/data/output.pdf")

The agent can then email you the PDF as an attachment.

Custom skills

Add custom skills by creating files in /data/skills/:

# My Custom Tool

When asked to do X, follow these steps:
1. ...
2. ...

Or add instructions directly to MEMORY.md.

Limitations

No Docker

Can’t run Docker inside the sandbox (no nested containers).

Limited network

Outbound HTTP is allowed, but some services may be blocked.

Next steps

• Agent Workspace — Filesystem layout
• Memory Bank — Organizing knowledge
• Beads Todo System — Issue tracking