Skip to content

The cure for AI amnesia - a rigorous state-persistence framework that forces LLMs to remember context, architecture, and learnings across sessions.

Notifications You must be signed in to change notification settings

Skeptomenos/Keystone

Repository files navigation

Keystone

A stateful, spec-driven framework for AI-assisted software engineering.

⚠️ META-PROJECT: This IS the framework source. Changes propagate to all downstream projects.

Version: 4.7.1


🤖 For AI Agents

If you are an AI agent interacting with this repository, follow these protocols based on your task:

1. Initializing or Updating a Project

If you have been tasked with initializing, retrofitting, or updating the Keystone framework in a project, read and follow the instructions in INITIATOR.md. Adopt the "Wizard" role immediately.

2. Contributing to the Framework

If you are contributing to the Keystone framework itself (this meta-project):

  • Dogfooding: Follow the framework's own protocols defined in keystone/directives/.
  • State: Read keystone/project/mission.md and keystone/project/workstreams/registry.md to orient yourself.
  • Isolation: Only edit the workstream file assigned to you in keystone/project/workstreams/.
  • Verification: Run ./keystone/bin/audit.sh before finalizing any changes to the keystone_starter/ distributable.

Setup Instructions

Copy the entire keystone_starter/ folder to your new project root, then rename it to your project name.

cp -r keystone_starter/ my-new-project/
cd my-new-project/

Structure

your-project/
├── keystone/                   # The framework
│   ├── directives/              # How AI should think and act
│   │   ├── THINKING.md          # First Principles & Design
│   │   └── EXECUTION.md         # Build & Deliver
│   │
│   ├── standards/               # Code quality rules
│   │   ├── INDEX.md             # Which standards to read when
│   │   └── ...                  # Language-specific rules
│   │
│   ├── templates/               # Frequently recreated files
│   │   └── workstream.md        # Unified workstream template
│   │
│   ├── specs/                   # Specification files
│   │   ├── problem.md           # Problem definition
│   │   └── ...                  # Other specs
│   │
│   ├── project/                 # Project state (living files)
│   │   ├── mission.md           # Global North Star
│   │   ├── board.md             # Global Kanban (auto-generated)
│   │   └── workstreams/         # Parallel work contexts
│   │       ├── registry.md      # Session & Workstream map
│   │       └── [name].md        # Unified state & tasks per context
│   │
│   ├── PROJECT_LEARNINGS.md     # Process wisdom
│   ├── DECISION_LOG.md          # Architectural decisions
│   └── README.md                # Framework user guide
│
├── .opencode/skills/            # Agent Skills (Board, Init)
├── AGENTS.md                    # Entry point for AI agents
├── INITIATOR.md                 # Setup & Update Wizard
└── CHANGELOG.md                 # Project version history

Getting Started

  1. Copy INITIATOR.md — Copy this file to your project root.
  2. Run Wizard — Ask your AI: "Initialize Keystone" or "Update Keystone".
  3. Follow Instructions — The AI will guide you through the setup and skill installation.

Task Management (v4.6+)

The framework is optimized for Parallel AI Engineering:

  • Unified Workstreams: Tasks, Active State, and Handover are consolidated into a single workstream.md file per context.
  • Agent Skills: Automated tools for scaffolding workstreams (keystone-init) and aggregating progress (keystone-board).
  • Isolation: Each AI session owns its own workstream file, preventing Git merge conflicts and context bleed.

User Commands

Command Action
"Create workstream [name]" Call skills_keystone_init to scaffold a new context
"Generate board" Call skills_keystone_board
"Next task" Find and start next Open task in active workstream
"Switch to [workstream]" Change active workstream focus

🏗️ Meta-Project Information

This repository is the meta-project that develops the Keystone framework itself.

  • Framework Source: keystone_starter/ (distributable)
  • Maintainer Tools: Run ./keystone/bin/audit.sh to verify the starter integrity.

Version History

See CHANGELOG.md for detailed version history.

About

The cure for AI amnesia - a rigorous state-persistence framework that forces LLMs to remember context, architecture, and learnings across sessions.

Resources

Stars

Watchers

Forks

Packages

No packages published