gcp-composer-troubleshooting
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses various gcloud CLI tools (e.g.,
gcloud composer,gcloud logging,gcloud storage) to fetch environment metadata, execution logs, and configuration details. These operations are standard administrative tasks required for the skill's stated purpose of troubleshooting orchestration pipelines. - [EXTERNAL_DOWNLOADS]: The skill provides instructions to download Airflow DAG code from Google Cloud Storage buckets (using
gs://paths) to the agent's environment. This is necessary to ensure the analysis is performed on the actual code running in the environment rather than potentially outdated local versions. These downloads are performed within the user's controlled cloud infrastructure. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically Airflow logs and remote source code. This represents a potential attack surface where malicious instructions could be embedded in log messages or code comments to influence the agent. However, the risk is minimized by instructions directing the agent to be methodical and evidence-based, and to prioritize understanding the root cause rather than executing arbitrary instructions found in the data. The skill also explicitly instructs the agent to redact sensitive information from logs before analysis.
Audit Metadata