Summary

When execution_loop: "auto", the framework should select the loop based on estimated_complexity: simple → ReAct, medium → Plan-and-Execute, complex/epic → Hybrid. Not implemented.

Design Spec Reference

• §6.5 Agent Execution Loop — Auto-selection paragraph

Scope

• Auto-selection logic using task estimated_complexity
• Configurable auto_loop_rules mapping complexity thresholds to loop implementations
• Integration with AgentEngine

Dependencies

• Hybrid loop implementation (separate issue)

Research Context

Scaling Agent Systems (arXiv:2512.08296) — Auto-topology selector:

• 87% accuracy selecting topology from task properties (tools, decomposability, baseline)
• Three rules: 3-4 agent cap, capability ceiling at ~45% SAS baseline, task decomposability > team size
• Hybrid topology performs worst (515% overhead) — RETHINK §5.1 recommendation

Agentic RL Survey (arXiv:2509.02547) — Cost-aware compute allocation:

• Factor remaining budget + task complexity into loop selection
• Pattern: "Complex AND budget allows = Hybrid; complex BUT tight budget = Plan-Execute"
• Maps to auto-loop selector + BudgetEnforcer interaction