Problem: Agents lose context between sessions, reusing AGENT.mds across projects is manual and can result in out of sync AGENT.mds across different projects as AGENT.mds are improved, project specific or generic reusable patterns are difficult to manager, there's no way to quickly install patterns from remote repositories.
I have been experimenting with ways to solve this problem and I ended up designing something I am calling the Agent Context Protocol, or MCP, which defines an AGENT.md and specifications for a specialized agent directory tree which includes templates for: commands, designs, milestones, and tasks.
Commands can be invoked by sending messages like `@acp.init` or `@acp.commit` for core ACP commands, or `@my-project.deploy` for project specific commands. Extension packages can be installed with a simple `@acp.package-install` ACP command.
Please consider checking out the README.md for a more advanced explanation of how ACP works, what problems it solves, and why it was created.
Problem: Agents lose context between sessions, reusing AGENT.mds across projects is manual and can result in out of sync AGENT.mds across different projects as AGENT.mds are improved, project specific or generic reusable patterns are difficult to manager, there's no way to quickly install patterns from remote repositories.
I have been experimenting with ways to solve this problem and I ended up designing something I am calling the Agent Context Protocol, or MCP, which defines an AGENT.md and specifications for a specialized agent directory tree which includes templates for: commands, designs, milestones, and tasks.
Commands can be invoked by sending messages like `@acp.init` or `@acp.commit` for core ACP commands, or `@my-project.deploy` for project specific commands. Extension packages can be installed with a simple `@acp.package-install` ACP command.
Please consider checking out the README.md for a more advanced explanation of how ACP works, what problems it solves, and why it was created.