╔══════════════════════════════════════════╗
║ ai-lib ║
║ Modular AI Workflow Library ║
╚══════════════════════════════════════════╝
A library of AI agent custom instructions. Some are half-baked, some are forged nicely. You may find something cool, or boring — like a library has.
Agent definitions, skills, behavioral rules, workflows, and MCP configs — all portable as an aipack.
flowchart LR
P[Pathfinder] -->|tasks.md| F[Forge]
F --> V[Validator]
V -->|qa-report.md - fail| F
V -->|pass| Done
| Layer | Count | What |
|---|---|---|
| Agents | 5 | Reusable personas with tool/skill bindings (Pathfinder, Forge, Validator, Orchestrator, Keeper) |
| Skills | 7 | On-demand instructions — load when needed |
| Rules | 5 | Always-on behavioral constraints (forge, pathfinder, validator, keeper, clean-code) |
| Workflows | 1 | Multi-step pipelines from idea to shipped |
| MCP | 2 | Portable server configs (github, playwright) |
# Prerequisite: Set your GitHub Personal Access Token
export GITHUB_PERSONAL_ACCESS_TOKEN=ghp_...
# Validator's browser E2E checks need a Playwright browser once per machine:
npx playwright install chromium
aipack pack install --url https://github.com/<your-user>/ai-lib.git --name ai-lib
aipack pack add ai-lib
aipack sync --harness all
# Then in any supported coding agent:
# /spark-to-steel — run the full pipeline- Machine-Readable — optimized for AI agents, not humans browsing
- Iteration-First — prove it works, then formalize it
- Portable — one source of truth, sync to any harness
See METADATA.md for the repository structure and agent protocol, and AGENTS.md for developer context.