The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the management of Airflow environments through the execution of CLI tools. This includes operational tasks such as triggering pipelines, clearing task runs, and discovering local instances via port scanning.\n
Evidence: Multiple commands in SKILL.md such as af runs trigger <dag_id>, astro dev start, and af instance discover --scan.\n- [COMMAND_EXECUTION]: A benign performance-optimization hook is executed upon skill initialization to warm the tool cache.\n
Evidence: The script hooks/warm-uvx-cache.sh executes af --version in the background when the session starts.\n- [PROMPT_INJECTION]: The skill ingests data from external Airflow environments, which constitutes a surface for indirect prompt injection attacks.\n
Ingestion points: Untrusted content enters the agent context via af tasks logs (execution logs) and af dags source (DAG source code) as documented in SKILL.md.\n
Boundary markers: Absent; the instructions do not currently wrap ingested data in delimiters or provide specific warnings to ignore embedded instructions.\n
Capability inventory: The agent can use the af CLI to trigger workflows (af runs trigger), modify configuration (af instance add), and access the Airflow REST API (af api).\n
Sanitization: Absent; no explicit filtering or validation of the ingested content is defined in the instructions before it is processed by the agent.

airflow