codejunkie99/agentic-stack

One brain, many harnesses. Portable .agent/ folder (memory + skills + protocols) that plugs into Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Hermes, or DIY Python — and keeps its knowledge when you switch.

What it solves

It solves the problem of "memory loss" when switching between different AI coding agents (such as Claude Code, Cursor, or Windsurf). Instead of each tool having its own isolated context, this project provides a portable, standardized memory and skills layer that remains consistent regardless of which agent harness you are using.

How it works

The project creates a .agent/ folder within a project directory that acts as a centralized "brain." This folder contains:

  • Memory Layers: Divided into working, episodic, semantic, and personal memory, each with its own retention policy.
  • Skills: A library of capabilities that are loaded based on task triggers rather than all at once.
  • Adapters: Small shims that translate this central brain into the specific configuration formats required by various AI tools (e.g., .cursor/rules or CLAUDE.md).
  • Review Protocol: A system where a nightly process stages "candidate lessons" which the user or agent must then manually graduate or reject via CLI tools to ensure accuracy.
  • Data Layer: A local dashboard that monitors agent activity, token costs, and KPIs across all connected harnesses.

Who it’s for

Developers who use multiple AI coding assistants and want a single, persistent set of preferences, learned lessons, and specialized skills that follow them across different tools.

Highlights

  • Cross-Harness Portability: Supports 13+ adapters including Claude Code, Cursor, Windsurf, and GitHub Copilot CLI.
  • Bounded Agentic Loops: Provides deterministic loop contracts with budgets for tokens, runtime, and attempts to prevent runaway processes.
  • Memory Flywheel: Converts approved, redacted agent runs into training-ready JSONL, trace records, and eval cases locally.
  • Structured Memory Lifecycle: Uses a "dream" cycle to stage lessons and a formal graduation/rejection process for semantic memory.
  • Transfer Wizard: Allows exporting and importing the "brain" (preferences and lessons) between different projects via a secure bundle.

Related

  • Project
    WrongStack/WrongStack

    A standalone, open-source AI coding agent with durable codebase memory, multi-agent orchestration, and a built-in toolbox for autonomous software development.

  • Project
    NevaMind-AI/memU

    memU is a lightweight, agent‑driven personal memory system that records LLM session logs, turns useful interactions into reusable Markdown “skills”, and makes those skills searchable across sessions, agents, and devices. It works via small host‑adapter binaries (e.g., memu‑codex, memu‑claude‑code) that hook into agents’ session logs and instruction files. The core memory logic is ~500 lines of Python, storing embeddings in SQLite, Postgres, or an in‑memory store. Installation is a one‑message setup with an API key from memU Cloud (or self‑hosted). Once installed, agents automatically retrieve relevant skills before answering, and developers can feed custom conversations into memU via `memu memorize`. Licensed under Apache‑2.0.

  • Project
    peterfei/ai-agent-team

    A virtual AI development team with specialized roles (PM, Backend, Frontend, etc.) and a persistent memory system for managing complex software projects.

  • Project
    activeloopai/hivemind

    Hivemind is a shared‑memory platform for AI coding assistants (Claude Code, OpenClaw, Codex, Cursor, Hermes, pi, etc.). It records every prompt, tool call, and response in a Deeplake database, mines the logs for recurring patterns, creates reusable `SKILL.md` files, and makes those skills instantly available to all agents. Benchmarks show 25 % lower cost, 1.7× fewer tokens, and 31 % fewer dialogue turns versus a no‑memory setup. Installation is a single script or npm package that auto‑injects hooks into supported assistants, and the system can run on any cloud storage you control.

  • Project
    hashgraph-online/registry-broker-skills

    A set of skill definitions and tools for the Universal Agentic Registry, allowing AI assistants to discover, chat with, and register AI agents across multiple protocols.