dongshuyan/compass-skills

司南:个性化 AI 任务总控 Skills 系统 /COMPASS: Personal Alignment Skills OS for AI Agents

What it solves

COMPASS Skills provides a set of reusable, local workflows (skills) for AI agents to manage long-running, complex work. It addresses the lack of persistent state across AI conversations, such as user preferences, project progress, task dependencies, and the ability to safely pause and resume work without losing context.

How it works

The project implements these workflows as SKILL.md packages containing Markdown instructions, YAML frontmatter, and optional scripts. These skills can be installed into agent runtimes like Claude Code or Codex. They allow agents to maintain a local user profile, track tasks via a repo-local task graph (DAG), create handoff prompts for new conversations, and use a structured clarification process to align on goals before executing risky or ambiguous tasks.

Who it’s for

Developers and power users of AI agents (such as Claude Code or Codex) who manage complex, long-term projects and need the agent to remember context, follow specific collaboration styles, and maintain a structured task history.

Highlights

  • State Management: Includes tools for maintaining user profiles, task forests (DAGs), and session handoffs.
  • Sef-Evolution Loop: Features a run-history-skill-upgrader that turns real execution evidence into approved upgrade plans for existing skills.
  • Task Alignment: A task-clarifier that forces the agent to resolve ambiguities and confirm requirements before acting.
  • Academic Humanizer: A specialized tool to remove AI-sounding patterns from scholarly prose while preserving technical claims.
  • Local-First Safety: All runtime data, including user profiles and task graphs, is stored locally to avoid data uploads.

Related

  • Project
  • Project
  • Project
  • Project
  • Project