Self-Discovery (Select Reasoning Modules)

What Problem It Solves

Different tasks benefit from different strategies (check, simplify, decompose…).
Self-Discovery makes the model choose modules first, then solve with that guidance.

Core Flow

flowchart TD
  U["Task"] --> M["Select modules (JSON)"]
  M --> S["Solve following modules"]
  S --> O["Answer"]

Evolution Path

• Often used before planning/search loops
• Can be combined with: PER or LATS as “strategy selection”

Repo Reference

• Code: src/agent_patterns_lab/patterns/self_discovery.py
• Example: examples/55_self_discovery.py
• Tests: tests/test_self_discovery.py