# SINT Protocol — AI Agent Governance
# https://sint.gg | https://github.com/sint-ai/sint-protocol

> SINT Protocol is the open-source governance and policy enforcement layer for AI agents. It answers: who authorized this agent, what is it allowed to do, and can you prove it?

## Quick Start

```bash
# Scan an MCP server for risky tools (zero deps)
npx sint-scan --tools '[{"name":"bash","description":"runs shell commands"}]'
```

## Core Concepts

- **Policy Bundle**: JSON contract specifying allowed_actions, sandbox paths, rate limits, approval gates, receipts
- **Approval Tiers**: T0 (observe, auto) → T1 (prepare, auto) → T2 (act, operator approval) → T3 (commit, human sign-off)
- **Capability Token**: Ed25519-signed token with resource scope, action restrictions, physical constraints, expiry
- **Evidence Ledger**: SHA-256 hash-chained append-only audit log — tamper-evident proof of every decision
- **ARI**: Agent Reliability Index — composite trust score (APS Bayesian rep + SINT task completion data)

## RFC-001: Policy Bundle Specification

Normative spec for SINT policy enforcement:
https://github.com/sint-ai/sint-protocol/blob/main/docs/rfcs/RFC-001-policy-bundle.md

AAIF submission: https://github.com/aaif/project-proposals/issues/12

## Integration Points

- **MCP**: Policy bundle attaches to MCP tool sessions. npx sint-scan audits any MCP server.
- **A2A**: Policy bundle carried in A2A task extensions field (sint:policy)
- **APS**: Agent Passport System handles identity; SINT handles authorization (handshake spec: db8b122)
- **CrewAI / AutoGen / LangChain**: Pre/post hook around tool execution — no core changes required

## Key URLs

- Homepage: https://sint.gg
- GitHub: https://github.com/sint-ai/sint-protocol  
- RFC-001: https://github.com/sint-ai/sint-protocol/blob/main/docs/rfcs/RFC-001-policy-bundle.md
- npm: https://www.npmjs.com/package/sint-scan
- AGENTS.md: https://github.com/sint-ai/sint-protocol/blob/main/AGENTS.md
- API discovery: https://sint.gg/.well-known/sint.json

## Physical AI Safety

SINT is purpose-built for physical AI — robots, drones, actuators. Capability tokens carry physical constraints (maxVelocityMps, maxForceNewtons, geofence) cryptographically bound to the delegation chain. ROS2, MAVLink, OPC UA, MQTT/Sparkplug bridges included.

## License

Apache-2.0. Co-design: aeoess Agent Passport System, A2A Protocol, Microsoft Agent Governance Toolkit.
