Back to directory

codex-deepseek-subagent

A Claude/Codex skill that configures DeepSeek as a native sub-agent inside a desktop coding app, letting the primary agent delegate text-only coding tasks to a chosen DeepSeek model using an official DeepSeek API key.

codex-deepseek-subagent

codex-deepseek-subagent is a skill for configuring and maintaining DeepSeek as a native sub-agent inside a desktop coding assistant app. It covers first-time setup, status checks, live testing, repairing the configuration after the parent model changes, and disabling or uninstalling it; it is not meant to be re-run for ordinary coding, exploration, or review tasks.

During setup, a user picks between a faster, cheaper model for everyday coding and a stronger model for more complex tasks, supplies an official DeepSeek API key through a local configuration page if one is not already stored, and the tool verifies the connection before marking the sub-agent ready. The primary agent can then delegate a task by spawning the DeepSeek sub-agent directly; DeepSeek only processes text, so the parent agent must first convert any image, video, or screenshot input into text.

API keys are stored in the operating system's credential store, and configuration changes are backed up before being written. It requires macOS or Windows and a compatible desktop coding app, and is MIT licensed.

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.