Roadmap: Agent Self-improving framework

[priyansh4320]

Description: On the framework level, support multiple strategies for agent self-improvement. This includes built-in mechanisms for evaluation and policies to determine what to evolve, when to evolve, how to evolve, and where to evolve (drawing on current research like A Survey of Self-Evolving Agents: On Path to Artificial Super Intelligence.
Strategies may include:

• Evaluation loops: Agent as a judge or HILT evaluation.
• Evolution triggers (when): on error, on drift detection, after fixed iterations, or by user feedback.
• Evolution scope (what/where): prompts, tool-use patterns, memory policies, orchestration logic, or model weights via agentic fine-tuning.
• Evolution methods (how): prompt rewriting, schema enforcement updates, tool routing, model fine-tuning, etc.