Back to directory

vibe-skill

A Claude Code skill that delegates coding subtasks to the cheaper Mistral Vibe agent, keeping Claude's own token usage to a small supervising role while it reviews the resulting diff.

vibe-skill

vibe-skill is a Claude Code skill that hands bounded coding tasks to Mistral Vibe, a separate coding agent, instead of having Claude read and edit files directly. When invoked, Claude decomposes the task and writes a prompt, a delegate process runs Vibe in a pseudo-terminal, and the delegate reports back its tool calls, token counts, and a git diff summary for Claude to review and summarize.

Because Vibe's own file reads and edits are not added to Claude's context, Claude's own token usage per run stays small and roughly constant regardless of how many files Vibe touches internally, while Vibe can be configured to use Mistral or another model or provider instead.

It is installed as a skill and script pair, and includes a written analysis comparing delegation cost against working directly, based on measurements the author collected from prior runs.

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.