Skip to main content

Skills

Skills are the foundation of the Pawz extensibility system — the first of three tiers that make Pawz infinitely customizable. Skills (this page) are prompt-only — they inject expert knowledge into your agent’s system prompt with zero configuration. For credential-based API/CLI connections, see Integrations. For custom sidebar views and persistent data, see Extensions.

Built-In Skills

Pawz ships with 400+ built-in integrations and skills that are compiled into the binary, extended to 25,000+ via the MCP Bridge to n8n. These include credential-based vault skills, CLI tools, API integrations, and instruction-based skills across 12 categories.

How built-in skills work

  1. Enable a skill in the Skills tab (its own sidebar view)
  2. Provide credentials (API keys, tokens) if required
  3. Install binaries (CLI tools) if required
  4. Enabled skills inject their tools and instructions into agent prompts automatically

Skill architecture

Skills are Rust functions registered in the skills/ module directory that the agent can invoke as tools. The architecture works as follows:
  • Registration: Each skill is defined as a SkillDefinition struct in skills/builtins.rs, specifying its ID, name, category, required credentials, tool names, required binaries, and agent instructions.
  • Per-agent toggle: Every skill is gated by an enable/disable toggle per-agent. When disabled, the skill’s tools and instructions are not injected into the agent’s prompt.
  • Categories: Skills are organized into 12 categories for the UI. Each skill belongs to exactly one category and can be independently toggled on or off.
  • Vault skills have dedicated tool functions (e.g. email_send, coinbase_trade) with encrypted credential storage. The engine executes these tools directly — credentials never appear in the agent prompt.
  • Instruction skills inject CLI/API knowledge into the agent’s system prompt. The agent then uses its built-in exec (shell) and fetch (HTTP) tools to interact with CLIs and APIs.
  • Credential encryption: Credentials are stored encrypted in SQLite. The encryption key lives in the OS keychain, separate from the database.

Built-in skill categories

Pawz ships with 400+ built-in integrations and skills across 12 categories:

Vault (Credential-based)

API Integrations

Productivity

Media

Image Gen uses the image_generate tool to create images from text prompts via Gemini (Google AI). Provide a detailed description including style, lighting, composition, and mood for best results.

Smart Home

Camera Capture (camsnap) can snap images from IP cameras, discover cameras on your network, and capture timed frame sequences. Use it for security monitoring, pet watching, or any vision-analysis workflow. Commands: camsnap snap <url>, camsnap discover, camsnap stream <url> --frames 10 --interval 1s.

Communication

CLI Tools

Development

System

Community Skills (Tier 1 — skills.sh ecosystem)

Community Skills are the first tier of the Pawz extensibility system — prompt-only knowledge packs that teach your agent new capabilities with zero configuration. They use the open-source skills.sh ecosystem. :::tip Graduating to higher tiers Need credentials, API keys, or dashboard widgets? Create a Tier 2 Integration. Need a custom sidebar view with persistent data? Create a Tier 3 Extension. Community Skills are the starting point — you can upgrade to higher tiers anytime. :::

How community skills work

Community skills use the SKILL.md format: a Markdown file with YAML frontmatter containing the skill’s name and description, followed by a markdown body with instructions for the agent. When installed, the instructions are injected alongside your built-in skills into every agent prompt — across all channels (WhatsApp, Telegram, Discord, etc).

Searching for skills

The Community Skills section sits at the top of the Skills view with a keyword search bar and popular category tags.
  1. Open the Skills tab in the sidebar — the Community Skills hero section is at the top
  2. Search by keyword: Type any term (e.g. “marketing”, “supabase”, “trading”) into the search bar and press Enter or click Search
  3. Pawz queries the skills.sh directory API (/api/search?q=) and displays matching skills with install counts (e.g. “22.9K”, “151.3K”)
  4. Popular tags: Click any of the quick-tag buttons (Coding, Marketing, DevOps, etc.) to instantly search that category
  5. Each skill card shows the skill name, install count, source repo, and a View on skills.sh link
No authentication is required — the skills.sh API is public.

Browsing a specific repo

You can also browse a specific GitHub repository directly:
  1. Scroll below the search section to the Browse Repository area
  2. Enter a GitHub repo in owner/repo format (e.g. vercel-labs/agent-skills)
  3. Pawz scans the repo tree for SKILL.md files and shows all available skills
  4. Click Install on individual skills or Install All to grab everything

Installing skills

  1. Click Install on any skill card — Pawz fetches the SKILL.md from the source GitHub repo
  2. The skill’s instructions are parsed from the Markdown body and stored locally
  3. Installed skills are automatically enabled and included in all agent prompts across every channel

Agent tools for skills

Agents have three built-in tools for working with community skills during conversations: This means you can ask your agent to find and install skills conversationally — no need to visit the Skills tab.

Managing community skills

  • Enable/disable: Toggle individual community skills on or off without removing them
  • Remove: Uninstall a community skill entirely (you can reinstall later)
  • Preview: Expand the instructions preview to see exactly what gets injected into the prompt

Compatible repositories

Any public GitHub repository containing SKILL.md files in the skills.sh format works. Browse the full directory at skills.sh to discover thousands of community-created skills with install counts and categories. You can also create your own skills repository — just add SKILL.md files anywhere in the repo tree.

Credential security

Credentials are stored in SQLite and encrypted with AES-256-GCM using a 32-byte random key stored in your unified OS keychain vault (service: openpawz). High-risk credentials (Coinbase, DEX) are server-side only — they are never injected into agent prompts. The engine uses them directly when executing trades.

Custom instructions

Each skill can have custom instructions that override the defaults. These are injected into the agent’s system prompt when the skill is enabled.

Smart prompt budgeting

When an agent has many skills enabled, their combined instructions can consume a large portion of the context window. The Rust engine manages this with a token budget system:
  1. Priority skills get full instructions — skills the agent is actively using or that match the current task
  2. Normal skills are compressed to ~300 characters — a compact reference the agent can still use
  3. Low-priority skills are dropped entirely if the budget is exhausted
This happens automatically before every agent turn. The result: agents have access to hundreds of tools without wasting context on irrelevant documentation. You never pay for token bloat from unused skill instructions. :::tip If you notice an agent forgetting how to use a specific tool, it may have been compressed or dropped due to budget limits. Either increase the agent’s context window or disable skills the agent doesn’t need. :::

Readiness check

The skill status indicator shows:
  • Ready — all credentials and binaries present
  • Missing credentials — API keys needed
  • Missing binary — CLI tool not installed
  • Disabled — skill is turned off