Claude Agent Tool

November 6, 2025 ยท View on GitHub

The claude-agent tool provides access to Anthropic's Claude large language models via the claude command-line tool. It allows you to use Claude as a sub-agent for various tasks, such as code generation, analysis, and troubleshooting.

Requires the Claude Code CLI to be installed and authenticated.

Parameters

  • prompt (string, required): A clear, concise prompt to send to the Claude CLI.
  • override-model (string, optional): Specify a different Claude model to use (e.g., opus). Defaults to sonnet.
  • yolo-mode (boolean, optional): Bypass all permission checks. Defaults to false.
  • continue-last-conversation (boolean, optional): Continue the most recent conversation. Defaults to false.
  • resume-specific-session (string, optional): Resume a conversation with a specific session ID.
  • include-directories (array of strings, optional): A list of additional directories to allow the tool to access.

Environment Variables

  • AGENT_TIMEOUT: (optional) The timeout in seconds for the claude command. Defaults to 300.
  • AGENT_MAX_RESPONSE_SIZE: (optional) Maximum response size in bytes. Defaults to 2097152 (2MB).
  • AGENT_PERMISSIONS_MODE: (optional) Controls whether yolo-mode parameter is exposed and its default behaviour. Options: yolo (force yolo-mode on, hide parameter), disabled/false (force yolo-mode off, hide parameter). If unset, agent can control yolo-mode via parameter. This controls the --dangerously-skip-permissions flag.
  • CLAUDE_SYSTEM_PROMPT: (optional) A string to append to the default system prompt.
  • CLAUDE_PERMISSION_MODE: (optional) Claude CLI's native permission mode setting (passed as --permission-mode to Claude CLI). This controls how Claude CLI prompts for permissions and is separate from AGENT_PERMISSIONS_MODE which controls --dangerously-skip-permissions.

Security Features

  • Response Size Limits: Configurable maximum response size prevents excessive memory usage and potential DoS conditions
  • Input Validation: Comprehensive parameter validation and type checking
  • Process Isolation: Agent execution runs in isolated subprocess with proper timeout controls
  • Timeout Controls: Configurable timeout limits prevent runaway processes
  • Error Handling: Secure error handling that doesn't expose sensitive system information