Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 

README.md

ADP Documentation

Specifications (v0.1)

# Document Description
01 Autonomy Taxonomy 5 standardized levels of agent autonomy (A1-A5) with governance requirements
02 Decision Classification 4 decision types x 4 risk levels x 3 reversibility classes
03 Trace Format Hash-chained immutable event schema for decision logging
04 Authorization Matrix Rules determining when human approval is required
05 Policy Schema Pre-built governance rule templates mapped to regulations
06 Regulatory Mapping Mappings to EU AI Act, Loi 25, NIST, ISO 42001, Singapore MGF, IEEE 7001
07 Agent Registry Structured inventory schema for AI agents in an organization
08 A2A Binding ADP as an A2A protocol extension: AgentCard declaration, task lifecycle, error mapping

JSON Schemas (v0.2)

Machine-readable validation schemas for all core specifications:

Schema Validates Spec
common.schema.json Shared definitions — 17 enums, 4 regex patterns All
trace-format.schema.json Decision trace events 03
policy-schema.schema.json Governance policies 05
agent-registry.schema.json Agent registry entries 07
authorization-matrix.schema.json Matrix configuration 04
a2a-binding.schema.json ADP metadata objects and AgentCard params 08

TypeScript Reference Implementation (v0.2)

The src/ directory contains a complete TypeScript SDK:

Module Description
src/types/ TypeScript interfaces matching all JSON schemas
src/core/ Core logic: classify(), authorize(), evaluatePolicy(), generateTraceId(), hashTrace(), verifyChain()
src/validators/ AJV-based schema validators for all document types
src/builders/ Fluent API builders: TraceEventBuilder, PolicyBuilder, AgentRegistryBuilder
src/__tests__/ 90 tests covering all matrix cells, overrides, edge cases, and the MCP server

Usage

import {
  classify, authorize, evaluatePolicy,
  TraceEventBuilder, PolicyBuilder, AgentRegistryBuilder,
  validateTrace, validatePolicy,
} from '@adp/core';

MCP Server (v0.3)

An MCP (Model Context Protocol) server wraps the SDK so any MCP-compatible AI assistant can call ADP governance directly. Run it with npx adp-mcp (or npm run mcp:dev).

Surface Items
Tools (8) adp_classify, adp_authorize, adp_evaluate_policy, adp_log_trace, adp_validate, adp_register_agent, adp_add_policy, adp_verify_chain
Resources (5) adp://registry, adp://policies, adp://traces, adp://matrix, adp://specs/{name}

Inputs are validated with Zod; session state lives in an in-memory store with SHA-256 hash-chained traces. See the main README for Claude Desktop setup.

Examples

See the examples/ directory for schema-validated integration examples.

White Paper

The founding document is available at gouvernance.ai/whitepaper.