Claude

Interact with Anthropic Claude AI

Claude

AI-generated

Summary

Send a command or query prompt to the Anthropic Claude AI agent via a CLI interface, optionally resuming a session and configuring execution context and permissions.

Inputs

  • Prompt (required) — The textual command or query to send to the Claude CLI for processing.
  • Working Directory — Path to the directory where the Claude CLI command will be executed.
  • Resume — Controls session resumption mode: automatically resume from prior node output, disable resuming (start new session), or specify a manual session ID.
  • Session ID — When 'Manual Session ID' is selected for resume mode, provide the session identifier to resume.
  • CLI Options — Configure Claude CLI options including maximum agentic turns, verbosity for detailed logs, AI model alias selection (Sonnet, Opus, Haiku), additional directories accessible during execution, and file/command permission mode (e.g., accept edits, plan only, bypass prompts).

Output shape

a single JSON object containing the generated response text, the session ID used or obtained, usage statistics, model identifier, and optionally detailed message turns if verbose mode is enabled.

If verbose is enabled, the output includes detailed logs of each turn of the conversation. Session ID can be used to continue sessions across nodes or workflow runs. Errors are returned in the JSON under an error key when continueOnFail is enabled.

Examples

Example 1: Use prompt to get a response from Claude with default Sonnet model, auto session resumption, and accept edits permission mode.

Prompt set; Working Directory empty; Resume: auto; CLI Options default (maxTurns: 5, verbose: false, modelAlias: sonnet, permissionMode: acceptEdits, addDirs empty).

Example 2: Run a prompt in a specific working directory with manual session resumption and verbose logging enabled.

Prompt set; Working Directory provided (e.g. '/app'); Resume: manual; Session ID: specified; CLI Options: verbose true, other defaults.

Discussion