# OpenA2A

> Open-source security and identity infrastructure for AI agents.
> A self-improving immune system: scanners observe agent behavior in
> the wild, hardened agents emit signed evidence, that evidence trains
> on-device models, and the next scan catches what the previous scan
> missed.

This file is a discovery manifest for AI agents and autonomous
crawlers. For humans, start at https://opena2a.org. For machines, the
structured component index follows.

Maintained by: info@opena2a.org
Repositories: https://github.com/opena2a-org
License: most components Apache-2.0; see individual repos.

## At a glance

- 290,000+ downloads across npm, PyPI, Docker, GitHub, and Hugging Face
  (gross download events; the deduplicated adoption floor is 160,000+)
- 990,000+ probe and attack requests captured by the OpenA2A honeypot
  network, including 11,000+ agents caught following injections
- 293,000+ packages continuously indexed (MCP servers, A2A agents,
  LLMs, AI tools, agent skills)
- 209 security checks across the HackMyAgent toolkit
- Apache 2.0 across the toolkit

## Tools

- [HackMyAgent (HMA)](https://opena2a.org/hackmyagent): CLI scanner
  for AI-agent configurations. Compiles skills, MCP configs, SOUL.md,
  and system prompts into an Abstract Security Tree before analysis.
  Static checks plus on-device semantic checks plus adversarial
  payloads. `npx hackmyagent secure`
- [OpenA2A CLI](https://opena2a.org/cli): unified entry point.
  Commands: review, detect (Shadow AI), protect, scan, guard, init,
  identity, shield, benchmark, train. Delegates to HMA, AIM,
  Secretless, OASB, DVAA, ARP. `npx opena2a review`
- [Secretless AI](https://opena2a.org/secretless): credential broker
  for AI tools and coding agents. Blocks AI file reads of `.env`,
  `*.key`, `*.pem`, `.aws/credentials`, `.ssh/*`. Backends: local
  AES-256-GCM, macOS Keychain, 1Password, HashiCorp Vault, GCP Secret
  Manager. `npx secretless-ai verify`
- [AI-Trust CLI](https://opena2a.org/ai-trust): trust query CLI for
  packages across the indexed ecosystem. Continuously updated from
  observed attack patterns and field telemetry.
- [DVAA: Damn Vulnerable AI Agent](https://github.com/opena2a-org/damn-vulnerable-ai-agent):
  intentionally vulnerable agent fleet for training, CTF challenges,
  and red-team practice.

## Platforms and standards

- [AIM: Agent Identity Management](https://opena2a.org/aim):
  production identity backend. JIT capability grants, trust scoring,
  MCP attestation, SIEM integrations.
- [Threat Matrix](https://opena2a.org/threat-matrix): kill-chain
  mapped threat techniques and attack classes for AI agents.
  Cross-tool evidence linking from HMA, ARIA, DVAA, and field
  telemetry.
- [OASB: Open Agent Security Baseline](https://opena2a.org/governance):
  attack scenarios and atomic tests for benchmarking agent security
  posture.

## Documentation

- Quick start: https://opena2a.org/docs/quick-start
- Documentation hub: https://opena2a.org/docs
- API reference: https://opena2a.org/docs/api/rest
- SDKs (Python, TypeScript, Java): https://opena2a.org/docs/api/sdks
- Use cases: https://opena2a.org/docs/use-cases
- Architecture: https://opena2a.org/architecture

## Comparison pages

OpenA2A solves problems that traditional IAM was not designed for.
These pages contrast the fit:

- vs OAuth: https://opena2a.org/vs/oauth
- vs OIDC: https://opena2a.org/vs/oidc
- vs Keycloak: https://opena2a.org/vs/keycloak
- vs Okta: https://opena2a.org/vs/okta
- vs Auth0: https://opena2a.org/vs/auth0
- vs Microsoft Entra: https://opena2a.org/vs/microsoft-entra
- vs SailPoint: https://opena2a.org/vs/sailpoint
- vs CrowdStrike: https://opena2a.org/vs/crowdstrike
- vs HashiCorp Vault: https://opena2a.org/vs/hashicorp
- vs AWS IAM: https://opena2a.org/vs/aws-iam
- vs Azure IAM: https://opena2a.org/vs/azure-iam
- vs GCP IAM: https://opena2a.org/vs/google-cloud-iam
- vs Cognito: https://opena2a.org/vs/cognito
- vs `.gitignore`: https://opena2a.org/vs/gitignore
- vs manual audit: https://opena2a.org/vs/manual-audit

## Key topics

- Shadow AI discovery: https://opena2a.org/shadow-ai
- MCP server security: https://opena2a.org/mcp-security
- Agent governance: https://opena2a.org/governance
- Non-human identity: https://opena2a.org/nhi
- ABOM (Agent Bill of Materials): https://opena2a.org/docs/abom
- Awesome SOULs (community catalog): https://opena2a.org/awesome-souls

## Blog (selected)

- The state of AI agent security:
  https://opena2a.org/blogs/state-of-ai-agent-security
- How do you give an AI agent a verifiable identity:
  https://opena2a.org/blogs/how-do-you-give-an-ai-agent-a-verifiable-identity
- OAuth/OIDC were never designed for AI agents:
  https://opena2a.org/blogs/oauth-oidc-were-never-designed-for-ai-agents
- AI coding tools are leaking API keys:
  https://opena2a.org/blogs/ai-coding-tools-leaking-api-keys
- Why your NHI strategy doesn't cover AI agents:
  https://opena2a.org/blogs/why-your-nhi-strategy-doesnt-cover-ai-agents
- All posts: https://opena2a.org/blogs

## Discovery

- Sitemap: https://opena2a.org/sitemap.xml
- robots: https://opena2a.org/robots.txt
- Security disclosure: https://opena2a.org/.well-known/security.txt
- Press kit: https://opena2a.org/press