AAK-A2A-001: Agent Card exposes internal capabilities
- Severity
- HIGH
- Category
- A2A Protocol (
a2a-protocol) - Scanner
agent_audit_kit/scanners/a2a_protocol.py- Auto-fixable
- No.
- SARIF name
A2aInternalCapabilities
What it detects
An A2A Agent Card exposes internal capabilities or admin-level skills that should not be advertised externally.
How to fix
Limit Agent Card capabilities to public-facing skills only.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-A2A-001
Limitations
Reads the capabilities entries only (a list of strings, or the keys of an object) and matches admin, system, internal, debug or root as a whole word: admin-console is reported, but admin_console, adminConsole and debugShell are not, and a skill such as admin-reset under skills is not looked at. Cards are found at agent-card.json and .well-known/agent.json in the project root and in any file named *agent-card*.json; a card at public/.well-known/agent.json is not found.
References
- OWASP Agentic Top 10: ASI07 (Insecure Inter-Agent Communication)
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-A2A-01 (A2A Internal Capabilities)
- CSA AI Controls Matrix (AICM): IAM-04
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-A2A-001.md: the page then shows that text, with the registry reference appended under it.