The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Airflow DAG logs and run details, which creates a surface for malicious instructions to be embedded in data.
Ingestion points: The skill retrieves external data using tools like get_dag_run_detail, analyse_dag_latest_run, and go_to_dag_log_view as described in SKILL.md.
Boundary markers: There are no explicit instructions provided to the agent to use delimiters or ignore instructions found within the logs or run metadata.
Capability inventory: The skill utilizes capabilities that can execute code or modify the system, specifically trigger_dag_run and the instruction to "Apply fixes in the DAG or related systems."
Sanitization: There is no evidence of data validation or sanitization before the retrieved log data is processed by the agent.

testing-dags