@robinpath/agent

AI agent integration for Claude Code and OpenAI Codex — prompts, parsing, caching, retries, batch processing, classification, extraction, guards, and context management

Why use this module?

The agent module lets you:

• Configure pipeline settings for AI agent execution
• Send a prompt to Claude Code CLI and parse the structured response
• Send a prompt to OpenAI Codex CLI and parse the structured response
• Set global debug verbosity level (0=off, 1=info, 2=verbose, 3=trace)
• Set the log file path for debug output

All functions are callable directly from RobinPath scripts with a simple, consistent API.

Installation

npm install @robinpath/agent

Quick Start

No credentials needed — start using it right away:

agent.claude "analyze" {"question": "What is 2+2?", "expectedOutput": "NUMBER"} into $answer

Available Functions

| Function | Description | |----------|-------------| | agent.pipeline | Configure pipeline settings for AI agent execution | | agent.claude | Send a prompt to Claude Code CLI and parse the structured response | | agent.codex | Send a prompt to OpenAI Codex CLI and parse the structured response | | agent.debug | Set global debug verbosity level (0=off, 1=info, 2=verbose, 3=trace) | | agent.log | Set the log file path for debug output | | agent.cost | Get pipeline cost and timing report for all executed steps | | agent.notify | Configure notification settings for pipeline events | | agent.model | Set or get the default AI model for all subsequent steps | | agent.prompt | Load a prompt template from a file with {{variable}} substitution | | agent.context | Manage conversation contexts for multi-turn AI interactions | | agent.batch | Process an array of items through an AI prompt with concurrency control | | agent.classify | Classify text into one of the given categories (sugar for common AI task) | | agent.extract | Extract structured fields from unstructured text as a JSON object | | agent.guard | Validate AI output against rules before passing it forward in the pipeline |

Examples

Send a prompt to Claude Code CLI and parse the structured response

agent.claude "analyze" {"question": "What is 2+2?", "expectedOutput": "NUMBER"} into $answer

Send a prompt to OpenAI Codex CLI and parse the structured response

agent.codex "generate" {"question": "Write a hello world in Python", "expectedOutput": "CODE"} into $code

Set global debug verbosity level (0=off, 1=info, 2=verbose, 3=trace)

agent.debug 1

Integration with RobinPath

import { RobinPath } from "@wiredwp/robinpath";
import Module from "@robinpath/agent";

const rp = new RobinPath();
rp.registerModule(Module.name, Module.functions);
rp.registerModuleMeta(Module.name, Module.functionMetadata);

const result = await rp.executeScript(`
  agent.claude "analyze" {"question": "What is 2+2?", "expectedOutput": "NUMBER"} into $answer
`);

Full API Reference

See MODULE.md for complete documentation including all parameters, return types, error handling, and advanced examples.

Related Modules

• @robinpath/openai — OpenAI module for complementary functionality
• @robinpath/anthropic — Anthropic module for complementary functionality
• @robinpath/ai — AI module for complementary functionality
• @robinpath/deepl — DeepL module for complementary functionality
• @robinpath/translate — Translate module for complementary functionality

License

MIT