Back to directory

COMPASS Skills

A set of nine local skills that give an AI coding agent a personal task-alignment system: clarifying goals before ambiguous work starts, tracking a repository-local task tree, and pausing and resuming work at safe checkpoints.

COMPASS Skills

COMPASS Skills is a collection of nine SKILL.md skills for AI coding agents, grouped into five runtime collaboration skills, two skill-engineering skills that learn from run history, one academic-writing skill, and one local hiring-support skill. The runtime skills align goals, scope, evidence and acceptance criteria before ambiguous or costly work begins; maintain a repository-local tree of goals, subtasks, dependencies and decisions; pause unfinished work at a safe boundary and resume it precisely later in the same conversation; compress a conversation's goal, progress and next steps into a prompt ready to paste into a new one; and keep a local, correctable profile of a person's communication and working preferences.

Two further skills turn a completed or repeatedly refined session into a new reusable skill package, or turn evidence from a real run into an upgrade plan for an existing skill that is only applied after explicit approval.

It installs selectively, skill by skill, through a package-style command for Claude Code, and is written in Python.

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.