skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/airflow-dag/Gen Agent Trust Hub
airflow-dag
Warn
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [DYNAMIC_EXECUTION]: The ConstructionDAGBuilder class in SKILL.md generates executable Python code for Apache Airflow DAGs. It uses string interpolation to embed user-provided parameters such as task_id, bash_command, and python_callable directly into the resulting script code.
- [COMMAND_EXECUTION]: The add_bash_task function allows users to define shell commands that are subsequently embedded in the generated Airflow DAGs and executed by the environment.
- [INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: Data for task definitions, bash commands, and DAG configurations are accepted from the user in SKILL.md. 2. Boundary markers: No delimiters or instructions are used to separate or ignore malicious content within user-supplied data. 3. Capability inventory: The skill performs file system write operations via the save_dag method to store generated scripts. 4. Sanitization: There is no input sanitization, escaping, or validation to prevent malicious code or commands from being included in the generated Python files.
- [PERSISTENCE]: The skill instructions suggest saving generated Python scripts to the /airflow/dags/ directory. Files in this location are automatically identified, loaded, and executed by the Apache Airflow scheduler, enabling persistent execution of any injected logic.
Audit Metadata