Built-in Skills

GolemBot ships with four built-in skills that are automatically copied into new assistant directories during golembot init or golembot onboard.

general — General Personal Assistant

A general-purpose skill that makes the agent a personal AI assistant.

Capabilities:

• Answer questions, provide suggestions, brainstorm ideas
• Read and write files: organize notes, generate reports, manage to-dos
• Execute scripts and commands for task automation
• Information retrieval and summarization

Persistent Memory: The general skill establishes a notes.md convention for cross-session memory — the agent reads it at conversation start and writes to it when important information comes up. See Memory for the full specification.

Skill Management: The agent can search and install community skills from ClawHub autonomously. When a user asks for capabilities the agent doesn't have, it proactively searches ClawHub and suggests installing relevant skills. All commands support --json for structured output.

Restrictions:

• Only operates within the assistant directory
• Does not modify golem.yaml, AGENTS.md, .golem/, or SKILL.md files

im-adapter — IM Response Guidelines

Optimizes the agent's responses for instant messaging platforms (Feishu, DingTalk, WeCom, Slack, Telegram, Discord).

Response length:

• Simple questions: 1–2 sentences, max 200 characters
• Complex questions: sectioned, each section max 300 characters
• Long content: key conclusion first, then offer detailed version

Formatting:

• Use standard Markdown syntax — it is automatically converted for each IM platform
• Use ## Heading for section titles, **bold** for emphasis, - item for lists
• Use fenced code blocks for code snippets and > quote for blockquotes
• Keep formatting clean: add blank lines between different block elements

Group chat:

• Address users by name; @mention them when helpful
• Extra concise to avoid flooding the chat
• In smart mode, output [PASS] when you have nothing important to add

Tone:

• Conversational and natural
• No overly formal greetings
• Never start with "Sure, let me help you with..."

multi-bot — Multi-Bot Collaboration

Enables coordination between multiple GolemBot instances running in the same fleet (on the same machine).

Capabilities:

• Peer awareness: The agent sees other bots in the fleet via [Peers: BotName (role)] injected into group prompts
• Domain-based response: When peers are present, the agent focuses on its own domain expertise and defers out-of-domain questions to the appropriate peer
• Cross-bot API access: The agent can call peer bots' HTTP API (POST /chat) for delegation when needed
• Fleet discovery: Peers are auto-discovered from ~/.golembot/fleet/ — no manual configuration required

How it works in group chats:

Group policy	Behavior with peers
mention-only	Responds when @mentioned; lighter guidance suggests deferring out-of-domain questions to peers
smart	Full [PASS] coordination — agent can stay silent when a peer is better suited to respond
always	Responds to every message; lighter guidance for peer awareness

Setup: This skill is installed automatically during golembot init. To enable effective multi-bot collaboration, set a persona.role for each bot (via golembot init --role "..." or in golem.yaml). The role is propagated to fleet registration so peers can see each other's specializations.

message-push — Proactive Message Sending

Enables the agent to send messages proactively to IM groups or individual users — without waiting for an incoming message.

Capabilities:

• Group messaging: Send messages to any group the bot is a member of
• DM messaging: Send direct messages to users who have interacted with the bot
• Channel discovery: Query available channels via GET /api/channels
• Intent recognition: Detects phrases like "send this to the ops group", "tell Alice that..."

API endpoints:

Method	Path	Description
POST	/api/send	Send a message to a specific channel + chatId
GET	/api/channels	List available channels and their send capability

Example:

curl -X POST http://localhost:3000/api/send \
  -H "Content-Type: application/json" \
  -d '{"channel": "feishu", "chatId": "oc_xxx", "text": "Meeting moved to 3pm"}'

Template Skills

The onboard wizard offers 6 scenario templates, each with a specialized skill:

Template	Skill	Key behavior
customer-support	faq-support	FAQ lookup from faq.md, escalation to unanswered.md
data-analyst	data-analysis	Reads from data/, outputs to reports/, uses calc.py
code-reviewer	code-review	5-dimension review, severity tiers (Must/Should/Nice), outputs to reviews/
ops-assistant	ops	Content writing, scheduling via schedule.md, competitor tracking
meeting-notes	meeting	Structured minutes, action items in action-items.md, archived to meetings/
research	research	Research reports with tables, sources in sources.md, archived to research/