Skip to content

Repository files navigation

CoWork-OS

CoWork-OS is a tool-agnostic Markdown control plane for agentic AI workflows.

It helps individuals and teams move from one-off prompting to structured AI collaboration with agents, roles, workflows, rules, memory, reviews, templates, adapters, examples, and handoffs.

CoWork-OS is not an app, not a SaaS product, and not tied to a specific AI tool. Codex, ChatGPT, Claude Code, Gemini, and other assistants are treated as adapters around a neutral workflow core.

Status

This repository is a clean-room public scaffold for agentic AI workflow documentation.

  • Private project data: not included
  • Real customer or operational data: not included
  • License: MIT, copyright CoWork-OS Contributors
  • Roadmap: see docs/roadmap-v0.1.md

Architecture

Core
-> Agents
-> Rules
-> Workflows
-> Prompts
-> Reviews
-> Templates
-> Adapters
-> Examples

Core Concepts

  • Core defines the neutral operating model.
  • Agents describe reusable roles and responsibilities.
  • Rules define non-negotiable safety and quality boundaries.
  • Workflows describe repeatable work patterns.
  • Prompts provide audit-ready task starters.
  • Reviews define quality gates before risky action.
  • Templates provide reusable project and handoff files.
  • Adapters describe tool capabilities without making them the core.
  • Examples show synthetic usage patterns.

Truth-State Priority

When sources disagree, use this order:

  1. Current user instruction
  2. Current repository state
  3. Repository-local rules and documentation
  4. Verified project record
  5. Handoffs and memory
  6. General model knowledge

Unverified information must be labeled as unverified.

Autonomy Gates

  • Default: read, analyze, plan, report.
  • Task Approval: edit files, create files, run non-trivial commands.
  • High-Risk Approval: commit, push, deploy, delete, publish, migrate, change infrastructure, or affect external systems.

When unsure, stop and ask for the smallest specific approval.

Quickstart

  1. Read README.md, docs/architecture.md, and rules/autonomy-gates.md.
  2. Start with prompts/audit-only-bootstrap.md.
  3. Produce a handoff using templates/HANDOFF.md.
  4. Run the relevant review gate before any high-risk action.

Example Use Cases

  • Audit a sample-api repository before editing.
  • Plan a demo-product feature with explicit approval gates.
  • Review an example-website change before release.
  • Organize research-notes into a verified handoff.

Not Included

  • Private projects
  • Private memory
  • Private handoffs
  • Local machine paths
  • Secrets or credentials
  • Tool-specific private configuration
  • Real customer, organization, or operational data

About

A tool-agnostic Markdown control plane for agentic AI workflows.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors