Back to directory

oberskills

A set of Claude Code and Codex skills for prompt engineering, subagent dispatch decisions and building new skills, backed by an MCP server that measures whether a skill's description actually triggers.

oberskills

oberskills is a set of skills aimed at making Claude Code and Codex better at tasks the author found they handled poorly: writing prose that reads naturally, searching the web without inventing URLs, building skills that work, and dispatching subagents sensibly. Its three main skills cover prompt design and adversarial review, guidance on when and how to dispatch a subagent, and creating, evaluating and reviewing Claude Code skills end to end.

A companion Bun and TypeScript MCP server spawns real headless Claude sessions to check a skill's frontmatter, structure and content against a specification, measure how often its description actually causes the skill to trigger, and iteratively optimize that description against a held-out test split. The same source tree supports both Claude Code and Codex through their respective plugin manifests.

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.