This commit introduces a new document, agentic_flow_patterns.md,
which details how system prompts (drawing examples from research.learn.md,
particularly Manus and Replit) define and manage agentic flow mechanisms.
The document covers:
- Key structural elements and tags (e.g., <agent_loop>, <event_stream>).
- Core patterns like the input-process-output loop and event-driven architectures.
- Task management, planning strategies, and policy integration.
- Conclusions on designing effective system prompts for robust agentic behavior.
