unified-ai-gateway
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Pulls a container image from the GitHub Container Registry (ghcr.io/happy520ai/unified-ai-system/mcp-server) using a pinned SHA256 digest to ensure the integrity of the downloaded content.- [COMMAND_EXECUTION]: Instructs the agent to execute multiple shell commands on the host environment to perform a security audit, which includes pulling Docker images, exporting filesystems, and searching for sensitive strings using recursive grep and find operations.- [COMMAND_EXECUTION]: Modifies host configuration by using the codex CLI to register a new MCP tool. The registration executes a Docker container configured with security-hardening flags, including --network none, --cap-drop ALL, and --security-opt no-new-privileges.- [INDIRECT_PROMPT_INJECTION]: The skill interacts with external tool outputs from the local gateway, which represents a potential surface for instructions embedded in tool responses to influence the agent's behavior.
- Ingestion points: Data returned by tools such as gateway_chat and gateway_health in SKILL.md.
- Boundary markers: The skill does not define specific delimiters or warnings to separate tool output from system instructions.
- Capability inventory: The agent possesses host-level execution capabilities through docker and codex CLI as part of the setup workflow.
- Sanitization: No explicit content validation or escaping is described for handling the responses generated by the gateway.
Audit Metadata