Agent Foundation¶

Specifications and foundational patterns for multi-agent systems.

Purpose¶

This repository contains stable specifications only — no implementation. It defines foundational patterns for:

Multi-agent system architecture and protocols
Agent identity, capabilities, and security models
Skill format, lifecycle, and hub protocols
Documentation standards and organization
Information flow tracking and taint analysis
Workspace bootstrap and setup procedures

Structure¶

/
├── agents/         # Agent definitions, manifests, registry protocols
├── skills/         # Skill definitions, formats, hub protocols
├── docs/           # Documentation definitions, formats, hub protocols
├── tools/          # Tool definitions, MCP protocol, execution policies
├── information-flow/ # Information flow tracking and taint analysis
├── bootstrap/      # Workspace setup and bootstrap guides
├── rules/          # Workspace rules best practices and templates
├── schemas/        # JSON schemas for validation (see schemas/README.md)

Getting Started¶

See bootstrap/development-environment.md for the full setup guide.

To bootstrap your workspace, paste this prompt into Amazon Q, Cursor, Zed, or any AI-enabled IDE:

Bootstrap my agent workspace following @agent-foundation/bootstrap/development-environment.md. Install agentctl, clone the required repositories, register the agent-skills hub, and install the agentctl skill.

Essential reading¶

bootstrap/development-environment.md — full workspace setup guide
rules/best-practices.md — how to write workspace rules for AI behavior
skills/authoring-guide.md — how to write and publish skills
docs/authoring-guide.md — documentation format and frontmatter standards
skills/hub.md — hub protocol and index format

Key Specifications¶

Mature specifications & Tooling¶

Skills — Skill format, lifecycle, hub protocols, authoring guide
Documentation — Document format, hub protocols, organization standards
Schemas — JSON schemas for validation (see schemas/README.md)
Bootstrap — Minimal setup procedures for agent workspaces
Rules — Workspace rules best practices and templates for AI behavior enforcement

Specification only¶

These specifications are stable but have no tooling implementation yet. They may benefit from further refinement as implementation experience is gained.

Agents — Agent definitions, manifest formats, registry protocols
Tools — Tool definitions, MCP protocol, execution policies
Information Flow Tracking — Taint tracking, TSL policy language, audit events

Tooling¶

agentctl is the CLI companion to this repository. It implements the hub protocols defined here:

Validate and generate index.json for skill and doc hubs
Manage hub registry (hub add/list/remove/enable/disable/refresh)
Install, update, and remove skills (skill install/list/update/remove)
Manage local config (config init/show/get/set)

brew tap geronimo-iia/agent && brew install agentctl
# or
cargo install agent-ctl

License¶

MIT

Contributing¶

See CONTRIBUTING.md.