agentic-stack Gives All Your AI Coding Tools One Shared Brain
agentic-stack is an open-source portable .agent/ folder that gives Claude Code, Cursor, Windsurf, and five other harnesses a shared memory and skills layer that survives tool switches.
Takeaways
- What: agentic-stack is a portable
.agent/folder (MIT-licensed) giving AI coding agents shared memory, skills, and protocols across 7+ harnesses. - Problem solved: Each AI coding tool (Claude Code, Cursor, Windsurf) has isolated memory; switching tools means starting from zero every time.
- Architecture: Four memory layers (working, episodic, semantic, personal) plus a nightly dream cycle that clusters episodic memories into reviewable lessons.
- Harness support: Full hook support for Claude Code and standalone Python; partial/manual for Cursor, Windsurf, OpenCode, Hermes, and OpenClaw.
- Skills: Five seed skills ship out of the box (skillforge, memory-manager, git-proxy, debug-investigator, deploy-checklist), each with a self-rewrite hook.
- Install: Free and open-source; available via Homebrew on macOS/Linux or PowerShell on Windows, with an onboarding wizard on first run.
Every AI coding assistant you use today is an amnesiac. Teach Claude Code that your project uses conventional commits, then open Cursor for a second opinion, and it has no idea. Using Cursor in your IDE, Claude Code in your terminal, Windsurf for second opinions? Each one has isolated memory. Teach one a rule, the others forget it. agentic-stack is a new open-source project that fixes this with a single portable .agent/ folder that works across all of them.
The problem is structural, not accidental
Teams using multiple AI coding tools are forced to duplicate project context across tool-specific formats like CLAUDE.md, .cursor/rules/, and .windsurfrules. There is no portable, tool-agnostic way to describe a project to AI agents. This is not a niche complaint. The agent framework coverage reflects how fragmented the agentic ecosystem remains. No single framework has won. Developers are building across all of them, and a memory layer that locks to one framework is a memory layer that developers will not adopt at scale.
The broader research community has noticed the same shift. The recent history of LLM agents can be understood as a progressive movement outward from the model itself. Capabilities were first treated as properties of weights, then as properties of prompts and context windows, and are now increasingly treated as properties of the broader infrastructure in which the model operates. agentic-stack is a direct bet on that last phase.
One brain, eight harnesses
The repo gives you the folder structure plus the files that actually go inside: a working portable brain with five seed skills, four memory layers, enforced permissions, a nightly staging cycle, host-agent review tools, and adapters for seven harnesses. The key design principle: the adapter layer is intentionally thin. For Claude Code, it generates CLAUDE.md and hooks configuration. For Cursor, it writes .mdc rule files. For standalone Python, it provides a