Vendor-agnostic AI context layer for the jayseGitHub monorepo.
Works with Claude Code, GitHub Copilot, Agentic, AndroidAgent, DesktopAgent, and any other AI tool — all reading the same plain-markdown source of truth.
ai-commons/
├── IDENTITY.md # Who I am, working principles, tech preferences
├── memory/ # Cross-project persistent memory (project decisions, constraints)
│ ├── MEMORY.md # Index
│ └── project_*.md # Per-project memory files
├── skills/ # Reusable AI skill definitions (vendor-neutral markdown)
├── rules/ # Coding conventions by stack
├── scripts/ # AI workflow scripts (ai-commit, ai-extract)
│ ├── ai-commit.py # Trace Claude Code sessions like Liquibase migrations
│ ├── ai-extract.py # Extract session metadata to YAML
│ └── install-hooks.py # Install git hooks for ai-commit
└── loaders/ # How to load this context from each AI tool
├── claude-code.md
└── copilot.md
- Plain markdown only — no proprietary formats, no tool-specific syntax
- Memory is the core value — project decisions, constraints, and WHYs that would otherwise be lost between sessions
- Loaders are thin — each AI tool has a small adapter; the content is shared
| Tool | Mechanism |
|---|---|
| Claude Code | additionalDirectories in settings.json + CLAUDE.md |
| GitHub Copilot | .github/copilot-instructions.md per repo |
| Agentic | WorkspaceBootstrapLoader global layer |
| AndroidAgent / DesktopAgent | AGENT.md + read_skill tool |