The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements Human-in-the-loop (HITL) operators that ingest untrusted input from human users via the Airflow UI or REST API.
Ingestion points: ApprovalOperator, HITLOperator, and HITLEntryOperator in SKILL.md capture user feedback and form data into params_input or chosen_options (e.g., in the entry_example and hitl_example code blocks).
Boundary markers: The instructions do not define delimiters or specific system prompts to isolate human-provided feedback from agent instructions.
Capability inventory: The skill demonstrates printing output to logs and performing task branching via HITLBranchOperator based on user input; these results are passed to downstream Python tasks (after_approval, process) which may possess broader execution capabilities.
Sanitization: The provided examples do not demonstrate sanitization, validation, or escaping of the params_input data before it is consumed by subsequent tasks.

airflow-hitl