| """ | |
| Purpose Agent Optimization — Epigenetic self-improvement pipeline. | |
| Optimize behavior BEFORE touching model weights: | |
| 1. Fingerprint capabilities from traces | |
| 2. Build filtered datasets from successful trajectories | |
| 3. Create prompt packs (optimized system prompts + skills + examples) | |
| 4. Shadow-evaluate candidates against baselines | |
| 5. Only if plateau persists: plan LoRA/distillation (optional) | |
| Key principle: prompt/skill/memory optimization first. Weight updates last. | |
| """ | |
| from purpose_agent.optimization.fingerprint import CapabilityFingerprint, fingerprint_traces | |
| from purpose_agent.optimization.dataset import TraceDatasetBuilder | |
| from purpose_agent.optimization.prompt_pack import PromptPack, PromptPackBuilder | |
| from purpose_agent.optimization.shadow_eval import ShadowEvaluator | |
| from purpose_agent.optimization.optimizer import AgenticOptimizer, OptimizationState | |
| __all__ = [ | |
| "CapabilityFingerprint", "fingerprint_traces", | |
| "TraceDatasetBuilder", | |
| "PromptPack", "PromptPackBuilder", | |
| "ShadowEvaluator", | |
| "AgenticOptimizer", "OptimizationState", | |
| ] | |