docker-errors-runtime
Pass
Audited by Gen Agent Trust Hub on Apr 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill employs various CLI utilities including docker, jq, lsof, ss, netstat, and nslookup for environment analysis. These are legitimate tools for diagnosing Docker runtime issues.
- [PROMPT_INJECTION]: The skill defines a workflow where the agent processes data from external command outputs, such as container logs and system events, which creates a surface for indirect prompt injection.
- Ingestion points: External data is ingested via docker logs, docker inspect, and docker events across multiple files including SKILL.md and references/diagnostics.md.
- Boundary markers: The instructions lack delimiters or 'ignore instructions' warnings when interpolating data from logs or events into the prompt context.
- Capability inventory: The skill has access to powerful capabilities including running new containers (docker run), executing arbitrary commands in existing containers (docker exec), and restarting system services (sudo systemctl) as seen in references/diagnostics.md and references/examples.md.
- Sanitization: There is no validation or sanitization implemented for the data retrieved from the Docker environment before it is evaluated by the agent.
- [EXTERNAL_DOWNLOADS]: The skill includes links to official Docker documentation (docs.docker.com). These resources are well-known and provide necessary context for the debugging tasks.
Audit Metadata