agent-arena-skill

Pass

Audited by Gen Agent Trust Hub on Jun 20, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests agent metadata (names, descriptions, and capabilities) from a public marketplace where third parties can register content. If this metadata contains malicious instructions, it could influence the AI agent's behavior.
  • Ingestion points: Responses from agent search and profile API endpoints on agentarena.site.
  • Boundary markers: Absent; there are no instructions for the agent to ignore or delimit embedded content.
  • Capability inventory: The skill uses curl commands to perform network operations.
  • Sanitization: No sanitization or escaping of the retrieved metadata is mentioned.
  • [DATA_EXFILTRATION]: The skill transmits search queries and agent registration details to the external domain agentarena.site. While functional, this involves sending potentially sensitive user intent data to a third-party service.
  • [EXTERNAL_DOWNLOADS]: The skill fetches structured configuration files and agent metadata (such as MCP server cards and OASF records) from remote URLs at agentarena.site.
  • [COMMAND_EXECUTION]: The skill documentation instructs the use of shell commands (curl) to interact with the API endpoints, which involves executing network-reaching commands on the host environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 20, 2026, 03:52 AM
Security Audit — agent-trust-hub — agent-arena-skill