Skip to main content
Part of the security infrastructure for AI agents

Star the project that secures AI agents.

OpenA2A ships open source security tools for the agents already running in production. Apache 2.0. No account required to try them.

$gh api -X PUT user/starred/opena2a-org/opena2a

Try first. Star after.

No signup. No config. Run the canonical commands and decide for yourself.

# Set up AI coding tools not to read credential files
$ npx secretless-ai init
# Run a security review on this project
$ npx opena2a-cli review
# Scan for vulnerabilities and attack payloads
$ npx hackmyagent secure

The repositories

Each repo solves a distinct part of the agent security problem. Star the ones you use.

opena2a

Unified CLI. One install for credential checks, governance scans, and identity onboarding.

hackmyagent

Security scanner for AI agents and MCP servers, with 318 static checks across 73 categories, 29 NanoMind semantic checks, and 164 adversarial payloads. The NanoMind checks run locally after a first-use model download from HuggingFace; the opt-in --deep analysis requires ANTHROPIC_API_KEY. OASB benchmarking and ARP runtime monitoring included.

secretless-ai

Stores API keys outside project files and sets up AI coding tools not to read credential files. In Claude Code a PreToolUse hook blocks credential file reads. For Cursor, Copilot, and Devin Desktop (formerly Windsurf) it writes an instruction file that nothing enforces.

agent-identity-management

Cryptographic identity, capability policies, trust scoring, and an audit trail for AI agents. Each agent signs a JWT with its registered Ed25519 key to get an access token, through the OAuth 2.0 JWT-bearer grant type of RFC 7523; the token endpoint returns no OpenID Connect ID Token.

damn-vulnerable-ai-agent

21 intentionally vulnerable agents, 12 vulnerability categories, 22 CTF challenges across 3 protocols (API, MCP and A2A). A training ground for red teams and defenders.

OASB

223 standardized attack scenarios for benchmarking AI agent security. Mapped to MITRE ATLAS.

Why your star matters

Open source security work is paid for in attention, not just commits. A star is the smallest unit of attention you can give, and it compounds.

Signal to other developers

Stars are how AI engineers discover security tools that work. A star tells the next person looking at the same problem that the project is real.

Funds open source maintenance

Stars surface our work to sponsors, conference reviewers, and standards bodies. That is how the tools stay free and Apache 2.0.

Shapes the roadmap

Repositories with traction get more issues, more pull requests, and faster fixes. Your star is a vote on what we focus on next.

What every repo gives you

The same standards across the org. No surprises in the package you star.

  • Apache 2.0 license. Permissive. Self host forever.
  • In the opena2a, hackmyagent, secretless-ai, damn-vulnerable-ai-agent and ai-trust tools, usage telemetry is on by default, carries no package names or scan content, and OPENA2A_TELEMETRY=off turns it off; the agent-identity-management SDKs send telemetry only after you turn it on, and OASB has none.
  • SLSA v1 provenance on npm for every hackmyagent version since 0.18.2, listed by npm view hackmyagent@0.33.2 dist.attestations
  • HackMyAgent 0.33.2 and opena2a-cli 0.10.13. Active release cadence.
  • Public roadmap. Decisions in the open via GitHub issues.
  • Security disclosure process for every repo. Coordinated, not punitive.

One click. Real signal.

If the tools have helped you, a star is the simplest way to give back.

$gh api -X PUT user/starred/opena2a-org/opena2a