Back to directory

Vibeweaver

A skill that enforces a strict process on a coding agent, such as decomposing a request, searching for prior art, designing before writing code, and verifying against runtime evidence, stopping the agent and making it start over when it drifts.

Vibeweaver

Vibeweaver is a skill, tuned mainly for opencode though meant to run on any agent harness, that does not try to make a model more capable; instead it tries to make the capability a model already has actually count, by enforcing a process: decompose the request, search for prior art, choose between at least two approaches, design before writing code, test first, verify against runtime evidence, and review before shipping.

The project's own published benchmarks show both directions: on the weakest model in its test set the full skill made results worse, because the model could not reliably follow the added process and the extra context made drift worse, while a more capable model reached a top score once the discipline was enforced. Enforcement is not just a prompt; a gate plugin and an auditor plugin watch the session and the files on disk for a missing required artifact.

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.