picoclaw-lite-agent

Add an entry to model_list in config.json with provider, model, and api_keys, or run picoclaw-lite model add with --api-base and --api-key for OpenAI-compatible endpoints. For openai, anthropic, and antigravity, picoclaw-lite auth login provides helper OAuth flows.

Use picoclaw-lite mcp add to register a server command, then verify with picoclaw-lite mcp list, show, and test. The CLI only manages tools.mcp.servers configuration; the gateway host loads and runs the servers at runtime.

Check that the directory name is a valid lowercase-or-hyphen skill name, that SKILL.md exists, and that frontmatter name and description are present. Also confirm the skill lives under workspace skills, global skills, or builtin roots, since loading follows that priority order.

Yes, PicoClaw supports ollama, lmstudio, vllm, and litellm through local OpenAI-compatible endpoints, typically without API keys. Add a model_list entry pointing at the local endpoint and set agents.defaults.model_name to match.

Gateway logs are written to ~/.picoclaw-lite/logs/gateway.log and gateway_panic.log, or under $PICOCLAW_HOME/logs if PICOCLAW_HOME is overridden. Run picoclaw-lite gateway --debug for verbose output, adding --no-truncate for full prompt and tool payload inspection.

Check whether PICOCLAW_CONFIG or PICOCLAW_HOME point to a different config file than the one you edited, and whether sensitive values belong in .security.yml instead of config.json. Also confirm whether you are testing the CLI path, the gateway path, or both.