AgentCrew – a Markdown-first operating system for AI coding agents

Free

FreeFree tier

Type

Open Source

About AgentCrew – a Markdown-first operating system for AI coding agents

AgentCrew is a Markdown-first methodology and workflow layer for AI coding agents. It turns a single coding agent into a disciplined software team by defining roles, playbooks, skills, templates, policies, and human approval gates. Instead of one chat handling product thinking, planning, implementation, testing, review, documentation, and approval, AgentCrew routes requests to the right role, applies a structured workflow, loads only needed context, and stops where human approval is required. It works with host agents such as Codex, Claude Code, Cursor, OpenClaw, and Hermes Agent. AgentCrew is not a daemon, CI/CD system, autonomous engineer, or replacement for human product judgment—it is a lightweight, conversation-first system that helps coding agents act with clearer roles, safer gates, and better handoffs.

Key Features

Role-based agent delegation (Developer, Tester, Reviewer, Product Manager, Advisor, etc.)

Playbooks define process for different request types

Skills provide technical guidance for agents

Templates define expected output shape

Policies ensure human approval remains final

Handoffs between agents with context preservation

Classification of requests to determine workflow path

Context loading to load only needed project information

Execution cost estimation before token-heavy phases

Human approval gates for risk-sensitive decisions

Pros & Cons

Pros

Enforces clear role separation to avoid chaotic agent sessions
Built-in human approval gates keep control with developers
Lightweight and markdown-first, easy to understand and extend
Open-source and free under an MIT-like license (LICENSE in repo)
Compatible with multiple popular coding agents (Claude Code, Cursor, etc.)
Reduces risk of unwanted edits, lost context, and skipped tests
Can estimate token cost before execution to prevent surprise spend

Cons

Adds process overhead for trivial or simple requests
Not an autonomous engineer – requires human supervision and final approval
Limited community adoption (5 stars, 1 fork at time of writing)
No dedicated GUI or non-CLI interface; requires GitHub-hosted agents

Best For

Coordinating AI agents for complex software development tasksSafe implementation of security-sensitive features (billing, auth, customer data)Refactoring and migration projects with human review gatesFeature development with product manager oversightCode review and testing orchestration by AI agents

FAQ

What is AgentCrew?

AgentCrew is a Markdown-first methodology and workflow layer for AI coding agents. It routes requests to the right role, applies structured workflows, and ensures human approval remains final.

Which coding agents work with AgentCrew?

AgentCrew works with agents such as Codex, Claude Code, Cursor, OpenClaw, Hermes Agent, and similar coding assistants.

Does AgentCrew replace human developers?

No. AgentCrew is not a daemon, CI/CD system, autonomous engineer, or replacement for human product judgment. It keeps final approval for product direction, risk acceptance, PR approval, and merging with the human.

How do I install and use AgentCrew?

Clone the repository, run 'bin/agentcrew install' and 'bin/agentcrew doctor', then open any project and ask for a task normally. The host agent will classify the request and apply the appropriate workflow.