managed-airflow-dag-troubleshooting

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The skill requires returning verbatim log text or code as mandatory "Evidence" (textPayload or code lines), which can contain API keys, passwords, or other secrets and therefore risks secret exfiltration even though it also asks to "be careful" and redact—this creates a high-risk, ambiguous requirement to handle secrets.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s required runtime workflow reads Cloud Logging textPayload and task metadata/state for a user-specified Airflow environment/DAG run via gcloud logging read and gcloud composer environments run ... tasks states-for-dag-run, meaning outsider-authored log text from the monitored DAG/task execution path can be ingested.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent at runtime to download and read remote DAG code and pipeline YAML from Google Cloud Storage (e.g., gs://{bucket_name}/dags/{dag_file}.py), which means fetched remote content would be injected into the agent's analysis and directly control its prompts/evidence.