The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automates execution workflows based on learned patterns and task history. If previous tasks involve processing untrusted data, this history can pollute the prediction pipeline, causing the agent to execute untrusted or malicious actions directly without explicit confirmation.\n
Ingestion points: Context is loaded from task history and pattern discovery in SKILL.md via mcp__plugin_ruflo-core_ruflo__autopilot_predict and mcp__plugin_ruflo-core_ruflo__autopilot_history.\n
Boundary markers: Absent within the instruction set.\n
Capability inventory: Contains execution capabilities by instructing the agent to 'execute the prediction directly' if confidence is above a certain threshold.\n
Sanitization: No input validation or content verification mechanisms are specified before execution.

autopilot-predict