factor_agent.llm.claude_code#
The Claude Code backend: claude -p headless, JSON output, every tool off, the local login.
Every built-in tool, MCP server, skill, slash command and settings file is
turned off and the call runs from an empty directory, so the model only
writes text and cannot read the data. ANTHROPIC_API_KEY and
ANTHROPIC_AUTH_TOKEN are kept out of the child’s environment, so the
call goes through the Claude Code login. Managed (enterprise) settings
cannot be turned off from the command line.
Classes
Complete a request with one headless Claude Code call. |