docker-manager
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE]: The skill includes commands to inspect container configurations, specifically targeting environment variables (e.g.,
docker inspect -f '{{.Config.Env}}'). This operation can reveal sensitive credentials, API keys, or connection strings stored within the container environment. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates reading container logs (e.g.,
docker logs -f). If a container outputs untrusted data or is compromised, an attacker could embed malicious instructions in the logs to influence the agent's behavior during log analysis. - Ingestion points: Container logs via
docker logs(SKILL.md) - Boundary markers: None identified in the provided instructions
- Capability inventory: Subprocess execution via
docker exec,docker start/stop, anddocker pull(SKILL.md) - Sanitization: None identified; logs are read directly from the Docker daemon
- [COMMAND_EXECUTION]: The skill provides the ability to execute arbitrary commands inside containers, including the use of the root user (e.g.,
docker exec -u root my-container /bin/bash). While powerful, this is a core functionality of Docker management.
Audit Metadata