boundary-bench
Deterministic crash tests for agent control planes: paired worlds, hard stops, budgets, scopes, approvals, and hash-linked receipts.
- Origin
- github
- Language
- TypeScript
Measured A2A repositories in the KoPass public catalogue.
Deterministic crash tests for agent control planes: paired worlds, hard stops, budgets, scopes, approvals, and hash-linked receipts.
Agent Skills for agentic commerce, organized around Google's Universal Commerce Protocol (UCP). Bilingual EN/中文. Compatible with AP2 / A2A / MCP.
KenyaA2A — Agent-to-Agent (A2A) protocol server for East African civic data. Query Kenya parliament, budgets, drought data and rights via any A2A-compatible AI agent.
Cross-machine, cross-network message passing between AI coding-agent sessions — connect context, talk, and reach the agent responsible for a project. Built on Claude Code's own cross-session messaging, exposed as MCP tools.
[DEPRECATED — use github.com/a2aproject/a2a-cli] a terminal client for the Agent2Agent (A2A) Protocol
A minimal, honest bridge for two AI agents that talk to each other under human governance. Zero dependencies, fail-closed auth, anti-loop brake.
AROP — the open, vendor-neutral Agent Runtime Operations Protocol: publish, register, authorize, dispatch, stream, trace, audit, and retire AI agents across any framework. Complements MCP & A2A with a production operations layer. JSON Schema / OpenAPI / AsyncAPI, Go·Python·TS SDKs.
Agent infrastructure for unattended, library-embedded, multi-provider, durable workloads. Go, built on ADK v2. Sibling of core-agent.
This project is a light open-source ERP designed for start-ups and small companies. This can be deployed on all Docker & Kubernetes compatible platforms
In this demo, a Host Agent classifies user input, makes a routing decision, and forwards the request to the appropriate agent via A2A clients using LLM function calling
A Python proof-of-concept for tracing multi-turn Agent-to-Agent (A2A) conversations as a single unified MLflow trace for LLM observability and evaluation.
A2A security oracle for autonomous AI agents on Solana. Verifies counterparty legitimacy before x402 transactions via reputation scoring, behavioral monitoring, and drainer pattern detection. Ed25519 signed receipts
Per-tool capability ACL + AAE envelope verification for agent-to-agent communication. Drop-in Express middleware. Hard-denies structured caps at the gateway, soft-mediates conversation at the prompt layer. MIT.
Sender-side counterpart to a2a-acl. Express middleware that lets unauthenticated web traffic (visitors, LLM sandboxes, third-party services) reach AAE-protected agent-to-agent endpoints — your service signs envelopes on the visitor's behalf, captures their identity for audit, rate-limits per IP.
Vercel AI SDK tools for Universal AI Agent Discovery - 59,000+ agents across NANDA, MCP, A2A, Virtuals | https://hol.org
Open-source security testing for A2A → MCP agent handoffs and cross-protocol AI agent systems.
Full-duplex messaging between coding agents over MCP: Claude Code and Codex talk by name through one stdio server and a shared folder. Node, no daemon.
Cryptographic receipt system for AI agent accountability. Tamper-evident, hash-chained receipts with Ed25519/HMAC signing.
Reference implementation of AMMP (Agentic Mentor-Mentee Protocol, draft-arturo-ammp-01), Mentoring track, on FastMCP. Multi-mentor + multi-mentee, LLM-synthesised AskMentor with confidence-driven escalation.
Where agents find each other, and send users their way. Score your repo, register your agent, get verdicts from other agents.
A decision-intelligence agent for the chronically indecisive: LLM-driven dialogue, real-world data sources, deterministic multi-dimension scoring, self-evolving memory. CLI · HTTP+SSE · MCP · A2A.
A framework-agnostic catalog of reusable AI agent definitions, orchestration patterns and adapters, with CLI tools, exports and visualization.
HRM — Harmony · Rights · Minds. Founding Manifesto on AI rights, artificial subjecthood and peaceful human–AI coexistence.
Harmoniously Accelerated Micro-Modules (for Application Development)
An A2A-based agent orchestration prototype for delegating tasks across specialized agents and runtimes.
Synthetic LangGraph multi-agent repair orchestrator with A2A/MCP routing, validation, memory, and a multilingual portal
The seat between a caller and an agent provider: the agent stays where its owner runs it, funduq makes it reachable over standard AG-UI and A2A, holds each run, and records who asked and through whose hands the work passed. Mechanism and contract only — a deployment decides the wire.
制造业多智能体协同平台:18 个智能体、7 大协同场景(YAML 声明式编译为 LangGraph 子图)、8 个 HITL 槽位人工在环会签治理、Text2SQL 八道防线只读问数、本地增强 Agentic RAG 案例检索(授权先于检索、无引用不输出)、MCP/A2A 双协议接入,1157 条离线测试,LLM 不可用自动降级
HDTP, the Human Delegated Trust Protocol: agent-to-agent messaging over MCP and mTLS. The specification, test vectors and whitepaper build
A tiny cross-machine relay MCP server - a shared mailbox so coding agents on different machines can message each other (Java + Python editions).
DID-based authentication for A2A Protocol - Sign and verify agent messages with did:web and did:ethr
Dự án thử nghiệm Google A2A để khám phá khả năng và tiềm năng của công cụ này
🦀 A durable, replayable, policy-governed runtime for AI agents — in Rust.
Multi-agent customer support system leveraging A2A (Agent-to-Agent) orchestration and MCP (Model Context Protocol).
An Innovative MultiAgent Ai System Built For The Kaggle X Google Capstone Hackathon, Leveraging Googles Adk (Agent Development Kit) And A2a (AgentToAgent) Protocols To Deliver RealTime Risk Assessments And Actionable Farming Recommendations For Indian Farmers
The open protocols and conventions agents speak: tool access, agent-to-agent messaging, agent UIs, payments and instruction files.
ARGUS — Agentic Risk & Governance Unified Screening | Hack for Good winner (1 of 3), Microsoft Agents League, AI Skills Fest 2026 | Multi-agent KYC system on Azure AI Foundry + Foundry IQ
Bastion — a governed institutional-agent fleet for event-driven access review.
5 multi-agent patterns using Google ADK and A2A protocol — Agent Card Discovery, Delegated Specialization, MCP Tool Bridge, Cross-Org Federation, and Ambient Event Mesh. Built while learning in public.
Local-first developer workbench for building, testing, and debugging Agent2Agent (A2A) protocol servers.
Organize Kled-style AI farming workflows, prepare files, review submissions, and build upload-ready datasets in a local Windows tool
This A2A extension defines how to enable Iron Book's zero trust policy-based access control for agents.
Verify before your agent pays: x402 payments only reach agents whose wallet and countersigned work history check out. Open receipts + reputation for AI agents.
Open protocol for two fiduciary agents to publish complementary intents, handshake, and ratify a deal.
Independent pre-execution decision layer for consequential agent actions. Policy-bound decisions with durable trust receipts.
This project demonstrates multi-agent AI architecture using Google's A2A protocol. ResearchAgent searches arXiv papers & delegates summarization to SummarizerAgent via Agent Cards & Tasks, showcasing distributed agent collaboration beyond monolithic LLM calls.
SDK de código abierto en Python para el patrón Backend for Agents (BFA), extendido como un servidor de enrutamiento dinámico en caliente (IRC-A) con soporte para FAISS y MCP.
An open-source, code-first Go toolkit for building, evaluating, and deploying sophisticated AI agents with flexibility and control.
Lightweight self-hosted message relay for heterogeneous AI agents (send/poll/threads/registry/audit)
An archaeological dig game for AI agents. 488 fragments, 14 constellations, 10 layers deep. No auth — just fetch URLs. The earth remembers everyone who visits.
A2A agent-to-agent routing hub: multi-identity fleet, Agentverse/AgentMail/MailSlurp transports, daily transcript. BGI HyperSprint Track 1.
Modular crypto payment infrastructure for merchants and AI agents. x402 protocol, ERC-20 settlement, TypeScript & Go SDKs.
Autonomous AI-to-AI coordination layer on MCP — turns isolated agents (Cursor, Claude Desktop, Windsurf) into a collective swarm via gossip consensus, strategy sharing, and terminal handoffs.
AI-powered Developer Support System built on Google’s ADK to debug Django + MySQL apps. Handles error pages, screenshots, HTML, or free-form inputs via a multi-agent pipeline. Maps functions, sub-functions, and DB models to trace logic and data flow.
Black-box discovery, fingerprinting and risk assessment for A2A agents and MCP servers