container-ops
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides commands that allow the agent to ingest untrusted data from Docker containers, which could contain malicious instructions.
- Ingestion points: Commands such as
docker logs,docker inspect, anddocker exec(found inSKILL.md) read data directly from containers or the Docker daemon. - Boundary markers: There are no boundary markers or instructions to ignore embedded content within the logs or metadata.
- Capability inventory: The skill includes powerful capabilities such as forced container deletion (
docker rm -f), resource modification (docker update), and file system manipulation (docker cp), which could be misused if an injection occurs. - Sanitization: No sanitization or validation logic is provided to filter or escape the output of the Docker commands before the agent processes them.
- [COMMAND_EXECUTION]: The skill is centered around executing shell commands for Docker management. While these are legitimate operations for its purpose, it provides the agent with a wide range of control over the container environment, including bulk deletion and resource updates.
Audit Metadata