The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze ComfyUI graph structures via panel_graph_outline and panel_query_graph. This data may originate from untrusted external workflow files. The agent is then encouraged to modify this environment using panel_set_widget, panel_connect, and panel_add_node, and to execute the results using panel_run. This pattern constitutes a standard attack surface where malicious instructions could be embedded in the graph metadata to influence the agent's behavior.
Ingestion points: Graph and node metadata retrieved in SKILL.md via panel_graph_outline and panel_query_graph calls.
Boundary markers: No specific boundary markers or instructions are provided to distinguish between workflow data and potential embedded instructions.
Capability inventory: The skill utilizes panel_add_node, panel_connect, panel_set_widget, panel_set_node_mode, and panel_run (execution of the workflow graph).
Sanitization: The instructions do not define sanitization, validation, or filtering for data retrieved from the ComfyUI nodes before the agent processes or acts upon them.

debug-render