ProDex icon

ProDex

Run OpenAI Codex, install skills, and manage your ProDex workspace. Complete ProDex Setup once before agent runs.

Actions6

Run Agent

AI-generated

Summary

Runs an OpenAI Codex agent on each incoming item using its required Prompt, with optional system instructions and skills. Returns the agent response, thread and run metadata; authentication comes from ProDex Setup on disk by default, or an explicitly selected ProDex Auth API credential when enabled.

Inputs

  • Use n8n Credentials — Off by default; reads authentication from ProDex Setup's auth.json. Turn on to use a selected ProDex Auth API credential. Complete ProDex Setup before running if using disk authentication.
  • Prompt (required) — Instruction sent to Codex. Defaults to the incoming item's chatInput, prompt, or text value.
  • System Prompt — Static instructions prepended to the task prompt.
  • Skills — Installed skills to apply. Selected skill instructions are added to the prompt.
  • Model — Codex model to run: GPT-5.5, GPT-5.4, or GPT-5.4 Mini; defaults to GPT-5.4.
  • Reasoning Effort — Reasoning level: extra high, high, low, or medium; defaults to medium.
  • Personality — Agent personality: default, friendly, or pragmatic.
  • Thread Mode — Start a new thread, continue the thread ID stored in node static data, or resume by ID. Defaults to a new thread.
  • Thread ID — Thread to resume when Thread Mode is Resume Thread By ID.
  • Sandbox — Codex filesystem access level: read only, workspace write, or full access. Defaults to read only.
  • Working Directory — Optional workspace directory for Codex.
  • Dynamic Skills — Per-run skills supplied as names, inline Markdown, or objects with a name and content.
  • Structured Output JSON Schema — Optional JSON Schema to constrain the agent's structured output.
  • Stream Progress To Execution Log — When enabled, logs completed agent messages to the execution log as progress.
  • Timeout (Seconds) — Run timeout in seconds; defaults to 300 (minimum 30).

Output shape

One output item per input item, each containing a JSON object with the agent result.

The object includes operation, output, threadId, items, usage, model, finishReason, and appliedSkills. Usage is null when unavailable; streaming mode returns the final streamed response with empty items and null usage. If Continue On Fail is enabled, failed items instead contain an error field.

Links

Discussion