en un clic
metaclaw-manager
// Self-management skill for MetaClaw - configure LLM models, channels, skills, and all settings through natural language
// Self-management skill for MetaClaw - configure LLM models, channels, skills, and all settings through natural language
| name | metaclaw-manager |
| description | Self-management skill for MetaClaw - configure LLM models, channels, skills, and all settings through natural language |
| license | MIT |
| compatibility | ["metaclaw"] |
| allowed-tools | ["ReadFile","WriteFile","EditFile","Bash"] |
| metadata | {"version":"0.1.0","author":"MetaClaw","builtin":true} |
This skill enables MetaClaw to manage its own configuration and state through natural language commands.
MetaClaw uses metaclaw.toml for configuration. The file is located in the project root or can be specified via --config.
To check current settings, read the metaclaw.toml file:
ReadFile: metaclaw.toml
Use EditFile to modify specific settings in metaclaw.toml. Common modifications:
Change LLM model:
[llm]
model = "claude" # Options: claude, gpt4, gemini, azure, huggingface, or full provider/model string
Adjust agent behavior:
[agent]
max_iterations = 25
sandbox = "basic" # Options: none, basic, docker
Enable/disable channels:
[channels.slack]
enabled = true
[channels.discord]
enabled = false
API keys and secrets are stored in .env (never in metaclaw.toml). Common variables:
ANTHROPIC_API_KEY - Anthropic ClaudeOPENAI_API_KEY - OpenAI GPTGEMINI_API_KEY - Google GeminiSLACK_BOT_TOKEN / SLACK_APP_TOKEN - SlackDISCORD_BOT_TOKEN - DiscordTELEGRAM_BOT_TOKEN - TelegramList installed skills:
metaclaw skill list
Install a skill:
metaclaw skill install <url-or-path>
Create a new skill:
metaclaw skill create <name> --desc "description"
Skills are stored in:
.metaclaw/skills/~/.metaclaw/skills/List channels:
metaclaw channel list
Start server with channels:
metaclaw start
Check system status:
metaclaw version
pip show metaclaw
Test LLM connection:
from metaclaw.llm import LLMProvider
provider = LLMProvider(model="claude")
# await provider.chat(messages=[{"role": "user", "content": "test"}])