Open-source security infrastructure for AI agents. Identity management, runtime protection, security scanning, compliance benchmarks, behavioral governance, and credential management. Apache-2.0, self-hostable, works independently or together.
npx opena2a-cli reviewOne command scans your project for shadow AI, credentials, config integrity, and governance gaps. Opens an HTML dashboard with a security score.
| Tool | What it does | Try it |
|---|---|---|
| opena2a-cli | Unified CLI. Shadow AI detection, identity, governance, scanning, protection. | npx opena2a-cli review |
| HackMyAgent | 209 static checks, 29 NanoMind semantic checks, 164 attack payloads, auto-fix. | npx hackmyagent secure |
| Secretless AI | Keep secrets out of AI tools: Claude Code, Cursor, Copilot, Windsurf. | npx secretless-ai init |
| AIM | Cryptographic identity, capability policies, audit logging, trust scoring. | opena2a identity create |
| ai-trust | Pre-install trust verification for AI packages: MCP servers, A2A agents, skills, AI tools. | npx ai-trust check <pkg> |
| Browser Guard | Detect and control AI agents in the browser. | Chrome Web Store |
| DVAA | Intentionally vulnerable AI agents for security training. | opena2a train start |
| Spec | What it defines |
|---|---|
| OASB-1 | 46 security controls across 10 categories, 3 maturity levels |
| OASB-2 | Behavioral governance — 72 controls across 9 domains, 4 agent tiers |
| OASB Eval | 222 attack scenarios for evaluating AI security tools |
| ABGS | Agent Behavioral Governance Specification — what goes in SOUL.md |
| awesome-agent-souls | 100+ SOUL.md templates by role, industry, and use case |
We contribute security fixes back to the projects we audit — not just reports, production code.
OpenClaw (205K+ stars) — 8 security PRs, 7 merged:
- Credential redaction in gateway config responses (#9858)
- Skill/plugin code safety scanner — 1,721 lines (#9806)
- Path traversal prevention in A2UI file serving (#10525)
- Timing-safe comparison for hook token auth (#10527)
- Supply chain hardening with --ignore-scripts (#10528)
- File permission enforcement for credential files (#10529)
- Security headers for gateway HTTP responses (#10526)
- Skill scanner false positive reduction (#10530)
┌─────────────────────────────────────────────────────────────────┐
│ opena2a-cli (unified entry point) │
│ npx opena2a-cli review │
├─────────────────────────────────────────────────────────────────┤
│ │
│ opena2a detect → Shadow AI discovery │
│ opena2a identity → AIM (cryptographic identity) │
│ opena2a scan → HackMyAgent (209 security checks) │
│ opena2a secrets → Secretless AI (credential protection) │
│ opena2a trust → ai-trust (pre-install trust check) │
│ opena2a mcp → MCP server signing and trust │
│ opena2a benchmark → OASB (222 attack scenarios) │
│ opena2a harden-soul → ABGS (behavioral governance) │
│ opena2a shield init → All of the above, one command │
│ │
└─────────────────────────────────────────────────────────────────┘
| Date | Update |
|---|---|
| Apr 17 | HackMyAgent v0.17.11 — detect Shadow AI audit command, --nanomind opt-in for AI-powered analysis, unified output across secure/scan-soul/harden-soul/explain, 209 checks + 29 semantic checks, 164 attack payloads across 16 categories |
| Apr 8 | opena2a-cli v0.8.19 — Runtime HMA version check, unified publish endpoint |
| Apr 8 | HackMyAgent v0.15.7 — GitHub repo scanning, unified publish endpoint |
| Mar 18 | opena2a-cli v0.8.7 -- OAuth login, identity lifecycle (suspend/revoke/reactivate), tag/MCP/activity management, server policies, detect --auto-scan |
| Mar 15 | opena2a-cli v0.7.2 — Shadow AI detection with governance scoring, HTML reports, CSV export, community trust data |
| Mar 15 | HackMyAgent v0.11.0 — Attack taxonomy, CVE-2026-25253 detection |
| Mar 14 | Secretless AI v0.12.4 — MCP server credential protection, broker auth hardening |
| Mar 11 | opena2a-cli v0.5.11 — Runtime command fixes, Shield improvements |
| Mar 10 | OASB-2 — Agent Soul specification published, 72 behavioral governance controls |
| Mar 5 | Secretless AI v0.11.4 — HashiCorp Vault backend, graceful fallback |
| Mar 4 | ABGS v1.0 — Agent Behavioral Governance Specification, 9 domains, 72 controls |
All tools are Apache-2.0.