Back to directory

Claude Dynamic Workflows — on Codex

A Claude Code skill that has Claude author a dynamic workflow script and run it across a fleet of local Codex GPT agents, streaming the run back as a live, interactive execution map.

Claude Dynamic Workflows — on Codex

Claude Dynamic Workflows — on Codex is a Claude Code skill invoked as a slash command: Claude Code writes a dynamic-workflow script describing agents, parallel steps, pipelines, phases and a budget, then runs it across many GPT-5 agents on a local Codex app-server instead of Claude subagents, streaming the run back as an interactive execution map.

The runtime holds the loop, branching and intermediate results so the calling context only sees the final answer, and workers can stay live: a worker can be steered on its warm thread, several can race with the losers cancelled, or a controller can adapt the plan as results land. A multi-workflow option lets Claude launch and supervise a whole fleet of concurrent workflows itself, answering their gates and steering them.

It ships as both the Claude Code skill and a standalone runner and viewer that works without Claude Code, including an offline, self-contained viewer with a bundled demo run. It is an unofficial community project, not affiliated with OpenAI or Anthropic.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.