codejunkie99/agentic-stack

⭐ 1,918  ·  Python  ·  GitHub Repo

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.

1-Sentence Summary

A portable .agent/ folder that persists memory and skills across 12+ AI coding assistants.

🔥 Key Capabilities & USP

• Portable Agent Brain (.agent/ folder): Solves the core pain point of losing all agent context, memory, and learned skills when switching between tools like Claude Code, Cursor, or Windsurf. Your agent’s knowledge travels with the project, not the tool.
• Harness Manager & Adapter System: A manifest-driven system (v0.9.0+) with a TUI and verb subcommands (add, remove, audit) lets you plug the same brain into multiple adapters simultaneously. No more re-training your agent for each IDE.
• Local Data Layer & Dashboard: Generates local dashboard exports (KPI summaries, token/cost estimates, daily reports) and a beta Mission Control web dashboard. All monitoring happens locally—zero external telemetry.
• Data Flywheel: Automatically exports approved, redacted runs into trace records, context cards, and training-ready JSONL artifacts. This creates a virtuous cycle of improving agent performance without calling external APIs or training models.
• Safe Upgrade & Manifest Repair: The upgrade and sync-manifest commands refresh skeleton-owned infrastructure without touching user memory, adapter settings, or existing skill directories. No fear of breaking your agent’s brain.

Technical Architecture

Component	Description
Language	Python backend for adapters, memory, tools, and skill management
Adapter System	Manifest-driven (adapter.json) with shared Python backend; supports 12+ harnesses
Memory Layers	Personal, semantic, episodic, and working memory; append-only audit history for lesson retraction
Cross-Platform	POSIX (macOS/Linux) via Homebrew; Windows via PowerShell; concurrent-write protection
Data Privacy	Local-first: all data and artifacts remain on your machine; no telemetry sent externally
Supported Harnesses	Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Copilot CLI, Gemini CLI, Hermes, Pi, Codex, Antigravity, DIY Python loops

Quick Start Guide

# macOS / Linux installation
brew tap codejunkie99/agentic-stack https://github.com/codejunkie99/agentic-stack
brew install agentic-stack

# Drop brain into a project (choose your harness)
cd your-project
agentic-stack claude-code
# or: cursor | windsurf | opencode | openclaw | copilot-cli | gemini | hermes | pi | codex | standalone-python | antigravity

# Windows PowerShell installation
git clone https://github.com/codejunkie99/agentic-stack.git
cd agentic-stack
.\install.ps1 claude-code C:\path\to\your-project

# Management commands
./install.sh dashboard           # TUI dashboard
./install.sh mission-control     # beta local web dashboard
./install.sh add cursor          # add a second adapter
./install.sh status              # one-screen view
./install.sh doctor              # read-only audit
./install.sh manage              # interactive TUI
./install.sh upgrade --dry-run   # preview safe upgrade
./install.sh upgrade --yes       # apply upgrade
./install.sh sync-manifest       # rebuild skill manifest
./install.sh remove cursor       # remove adapter

# Mission Control beta
agentic-stack mission-control --port 8787
agentic-stack mission-control --snapshot

# Lesson retraction
.agent/tools/retract_lesson.py <lesson_id> --rationale "..."

Pros, Cons & Use Cases

Pros

• Eliminates context loss when switching between coding agent tools—your agent remembers everything.
• Local-first architecture means no external telemetry or API calls for core functionality; privacy is baked in.
• Broad adapter support (12+ harnesses) with easy add/remove management via TUI.
• Safe upgrade process that preserves user data, settings, and skill directories.
• Built-in monitoring with dashboard, data flywheel, and artifact generation capabilities.

Cons

• tldraw visual canvas skill is beta and opt-in (disabled by default).
• Requires initial installation via Homebrew (macOS/Linux) or PowerShell (Windows); no pip install yet.
• Pre-v0.9 users must run doctor first to migrate to the new manifest system—a one-time migration step.

Who should NOT use this?

• Single-tool purists who only ever use one coding assistant and never switch—the core value proposition is wasted.
• Teams requiring cloud-hosted agent memory for centralized compliance or audit trails—this is strictly local-first.
• Users allergic to CLI tooling who prefer GUI-only configuration—the installation and management are terminal-based.

Ideal Use Cases

• Polyglot developers who switch between Claude Code, Cursor, and Windsurf depending on the task or project phase.
• Teams experimenting with multiple AI coding assistants to find the best fit, without losing agent knowledge.
• Privacy-conscious engineers who want full control over their agent’s memory and telemetry data.
• Power users building custom agent harnesses (DIY Python loops) who want a standardized memory and skills layer.

Community & Activity

With 1,918 stars and active development through May 2026, codejunkie99/agentic-stack has clearly struck a nerve with the developer community. The project is solving a real, painful problem that every multi-tool AI user has experienced: context loss when switching assistants. The momentum is strong, the architecture is well-thought-out, and the recent v0.9.0 manifest system upgrade shows a team committed to clean, safe evolution. This is a project to watch—and adopt—if you live in the multi-harness AI coding world.