​

Skills Overview

OrcBot provides 40+ production-ready skills that enable autonomous reasoning, multi-channel communication, web browsing, system command execution, and multi-agent orchestration. Skills are the core building blocks that power OrcBot’s autonomous capabilities.

​

Skill Architecture

Every skill in OrcBot includes:

• Name: Unique identifier used by the LLM to invoke the skill
• Description: Natural language description that appears in the agent’s context
• Handler: JavaScript/TypeScript function that executes the skill’s logic
• Metadata flags: Control orchestration behavior (isDeep, isResearch, isSideEffect, etc.)

​

Skill Categories

​

🌐 Web & Browser Skills

High-level navigation and research skills:

• web_search - Multi-provider search with API + browser fallback
• browser_navigate - Navigate to URLs and extract semantic snapshots
• http_fetch - Lightweight HTTP requests without launching a browser
• extract_article - Clean article extraction via Readability
• download_file - Download files with 50MB cap and timeout protection

See Web Search for detailed documentation. Low-level browser automation:

• browser_examine_page - Get semantic snapshot of current page
• browser_click - Click elements by selector or reference
• browser_type - Type into form fields
• browser_wait_for - Wait for elements to appear
• browser_screenshot - Capture screenshots
• browser_vision - AI-powered visual analysis
• browser_solve_captcha - Automatic CAPTCHA solving
• browser_run_js - Execute custom JavaScript
• switch_browser_profile - Manage persistent browser profiles
• switch_browser_engine - Toggle between Puppeteer and Lightpanda

See Browser Tooling for detailed documentation.

​

📁 File System Skills

• read_file - Read files with line range support (start_line/end_line)
• write_file - Write or append to files with 10MB cap
• create_directory - Create directories recursively
• list_directory - List directory contents
• download_file - Download from web to local storage

See File Operations for detailed documentation.

​

🐚 Shell & Execution Skills

• run_command - Execute shell commands (PowerShell on Windows)
• execute_typescript - Compile and run TypeScript snippets
• execute_python_code - Run Python in isolated virtual environment
• install_npm_dependency - Install Node packages for custom skills
• install_python_package - Install Python packages via pip
• get_system_info - Get OS, platform, and shell info
• system_check - Verify commands, libraries, and paths exist

See Shell Execution for detailed documentation.

​

🧠 Memory & Learning Skills

• recall_memory - Semantic search across all memory types
• update_user_profile - Persist user preferences and facts
• update_learning - Research topics and save to LEARNING.md
• update_journal - Write reflections to JOURNAL.md
• update_agent_identity - Evolve agent personality
• update_world - Maintain agentic society rules in WORLD.md
• deep_reason - Intensive chain-of-thought analysis
• search_chat_history - Search conversation history
• get_whatsapp_context - Get contact profile + recent history

RAG Knowledge Store:

• rag_ingest - Ingest documents into vector knowledge base
• rag_ingest_file - Ingest local files
• rag_ingest_url - Download and ingest web content
• rag_search - Semantic search across ingested knowledge
• rag_list - List documents and collections
• rag_delete - Delete documents or collections

See Memory & Learning for detailed documentation.

​

💬 Channel Messaging Skills

Core messaging:

• send_telegram - Send Telegram messages
• send_whatsapp - Send WhatsApp messages
• send_discord - Send Discord messages
• send_slack - Send Slack messages
• send_email - Send emails via SMTP
• send_gateway_chat - Send to web gateway interface

Rich interactions:

• send_file - Send files (auto-detects channel)
• send_voice_note - TTS + voice message delivery
• send_image - Generate AI image and send in one step
• telegram_send_buttons - Inline keyboard buttons
• telegram_send_poll - Native polls
• telegram_edit_message - Edit sent messages
• telegram_pin_message - Pin important messages
• react - React with emojis (auto-detects channel)

Email management:

• search_emails - Search inbox with filters
• fetch_email - Get full email by UID
• index_emails_to_knowledge_base - Ingest emails for semantic search
• generate_email_report - Synthesize multi-email reports

See Channel Messaging for detailed documentation.

​

⏰ Scheduling Skills

• schedule_task - One-time task scheduling (relative or cron)
• heartbeat_schedule - Recurring autonomous tasks
• Polling system - Event-driven condition monitoring

See Scheduling for detailed documentation.

​

🤖 Multi-Agent Orchestration

• spawn_agent - Create sub-agents for parallel work
• create_peer_agent - Create independent peer agents
• delegate_task - Assign tasks to agents
• distribute_tasks - Auto-assign pending tasks
• orchestrator_status - Get orchestration summary
• complete_delegated_task - Mark task completed
• send_agent_message - Inter-agent messaging
• broadcast_to_agents - Broadcast to all agents

See Orchestration for detailed documentation.

​

🖥️ Computer Use (Vision + System)

• computer_screenshot - Capture browser or system screenshots
• computer_click - Click by coordinates or vision-locate
• computer_type - Type with optional vision-guided input location
• computer_key - Press keys or key combos
• computer_mouse_move - Move mouse cursor
• computer_drag - Drag operations
• computer_scroll - Scroll in browser or system
• computer_locate - Vision-locate elements by description
• computer_describe - AI vision description of screen regions

​

⚙️ Configuration & System Management

• manage_config - Agent-driven configuration with policy-based security
• create_custom_skill - Create executable TypeScript plugin skills
• self_repair_skill - Diagnose and fix failing plugins
• tweak_skill - Patch built-in or plugin skills
• manage_skills - Append skill definitions to SKILLS.md

Self-tuning:

• tune_browser_domain - Adjust browser settings per domain
• mark_headful - Mark domains requiring visible browser
• tune_workflow - Adjust step limits and retry settings
• get_tuning_state - View current tuning configuration
• reset_tuning - Reset to defaults

​

Skill Metadata Flags

Skills use metadata to drive orchestration:

​

Best Practices

​

Error Handling

All skill handlers return string results or error messages. Common error patterns:

• Missing parameters: Error: Missing [parameter]. [helpful guidance]
• Unauthorized operations: Error: Only admin users can [action]. [reason]
• Channel not available: [Channel] not available. Configure [requirement] first.
• File not found: Error: File not found: [path]. [suggestions]

The decision engine analyzes errors and automatically retries with corrected parameters or alternative strategies.

​

Dynamic Plugin System

In addition to core skills, OrcBot supports:

• Hot-loadable TypeScript/JavaScript plugins from ~/.orcbot/plugins/
• Agent Skills (SKILL.md) following the agentskills.io specification
• Self-repair - Broken plugins are automatically diagnosed and fixed
• Progressive disclosure - Only activated skills load full instructions

Plugins inherit the same metadata flags and orchestration behavior as core skills.

​

Next Steps