Agent-Friendly Standards & RFCs
AFN publishes formal technical specifications, schema requirements, and protocol conventions to ensure tools, CLIs, and web services are reliably operable by autonomous AI agents.
Initial community proposal submitted via GitHub PR. Open for technical debate and prototyping.
Reviewed by Working Group. Requires at least two independent reference implementations.
Approved by AFN Technical Steering Committee. Finalized standard baseline.
Ecosystem architectural reference and interoperability documentation.
Active Specifications & RFCs
TOTAL: 8 SPECIFICATIONSAgent-Friendly Command Line Interface Specification
Defines runtime flags, machine-readable output schemas (--json, --schema), deterministic exit codes, and non-interactive execution guarantees for terminal utilities.
- -Strict structured JSON/JSONL output over stdout with zero terminal escape noise
- -Dynamic command discovery via --schema and tool definition introspection
- -Deterministic non-zero exit codes mapped to standardized error categories
- -Non-interactive fallback guarantees preventing unhandled STDIN prompt deadlocks
Deterministic API Contracts for Autonomous Agents
Establishes idempotency keys, rate-limit discovery headers, machine-readable error recovery hints, and atomic operation boundaries for Web APIs.
- -Mandatory Idempotency-Key handling on all mutating state endpoints
- -Granular Retry-After and rate-limit discovery headers
- -Machine-actionable error payloads conforming to RFC 9457
- -OpenAPI 3.1 schema publishing for autonomous tool discovery
Agent-Friendly Web Property & Documentation Specification
Specifies the llms.txt standard files, markdown content negotiation, and accessibility tree preservation for indexing website capabilities.
- -Plaintext semantic context mapping via llms.txt and llms-full.txt
- -Content negotiation supporting Accept: text/markdown for clean LLM ingestion
- -Explicit robots.txt AI agent policies and security declarations
- -Semantic HTML tags and accessible ARIA landmark preservation
Agent-Friendly Model Context Protocol (MCP) Server & Tool Specification
Establishes production-grade design and governance guidelines for Model Context Protocol servers, tools, timeouts, and token budgeting.
- -Concise, intent-driven tool naming and descriptions preventing model distraction
- -Strict JSON Schema input validation with explicit enumerations and bounds
- -Mandatory tool execution timeouts (max 30s) and cancellation token handling
- -Token-optimized payloads with automatic pagination and summarization
Agent-Friendly Structured Data & Knowledge Specification
Specifies dataset publication, metadata indexing, schema declaration, and entity resolution designed for direct ingestion by AI agents.
- -Explicit schema definitions via JSON Schema, JSON-LD, or Parquet
- -Standard global entity URIs and identifier resolution
- -Dataset manifests with pre-computed semantic chunks for RAG pipelines
- -Semantic versioning and immutable snapshot hashing for deterministic benchmarks
Agent Communication Protocol (ACP 2.0) Specification
Defines structured task dispatch, execution lifecycle management, and economic escrow settlement between Clients, Gateways, and Agent Services.
- -Authoritative task-layer protocol superseding AFN-002 transport abstractions
- -Standard lifecycle state machine: created -> pending_escrow -> running -> delivered -> settled
- -Cryptographic request envelopes with idempotency tokens and token budgeting
- -Deterministic dispute arbitration interfaces and per-result pricing support
Agent Work Provenance and Reputation Attestation Specification
Defines verifiable acceptance receipts and cryptographic work provenance attestations for agent deliverables, preventing Sybil reputation inflation.
- -Cryptographically signed AcceptanceReceipt primitives binding client, agent, and deliverable SHA-256 hashes
- -Mathematical multi-factor reputation scoring (completion rate, accuracy, repeat hire)
- -Automated provenance ledger integration for agent marketplaces
- -Sybil resistance through verified financial escrow settlement proof
Agent-Friendly Interface Standard Reference (AFX)
Community reference architecture for standardizing agent interface abstractions, state representations, and execution wrappers across heterogeneous tools.
- -Community reference for agent interface conventions
- -Unified abstraction layer over legacy command-line tools
- -Bi-directional event streaming and execution tracing primitives
Technical Working Groups
Chartered committees responsible for drafting and maintaining RFC domain tracks.
CLI Automation Working Group
Standardizing command-line interfaces for reliable, structured, deterministic AI Agent execution.
API & Handshake Working Group
Defining robust, self-remediating, and deterministic API protocols tailored for autonomous systems.
Web Discoverability Working Group
Establishing web-scale discoverability standards (llms.txt, semantic negotiation) for machine agents.
MCP & Context Working Group
Formulating standards for Model Context Protocol servers, tool safety, timeouts, and schema clarity.
Data & Semantics Working Group
Defining machine-readable dataset packaging, semantic chunking, and JSON-LD entity linking standards.
Agent Protocols & Settlement Working Group
Formulating structured task dispatch, execution lifecycles, and economic escrow settlement standards across agent networks.
Reputation & Attestation Working Group
Establishing cryptographic work receipts, deliverable provenance, and verifiable trust metrics for autonomous agents.
Propose a New Standard (RFC)
Have an interface pattern, CLI schema, or agent discoverability mechanism? Submit an RFC proposal to the AFN repository.