Lifting Traces to Logic — programmatic skill induction with neuro-symbolic learning targets long-horizon agentic tasks

A new arXiv paper titled 'Lifting Traces to Logic: Programmatic Skill Induction with Neuro-Symbolic Learning for Long-Horizon Agentic Tasks' proposes a methodology for extracting reusable program-like skills from neural reasoning traces and re-using them across agentic workflows. The result is a step toward closing the gap between transformer-style reasoning (broad but expensive) and symbolic planning (narrow but cheap).

The methodology pattern is the interesting move. Traces from successful agentic task completions get lifted into logical predicates, the predicates accumulate into a skill library, and future agentic queries can call into the library directly rather than reasoning from scratch. For multi-step agentic workflows that share substructure, the inference cost reduction is substantial.

For the sparse-policy-selection finding, this is a complementary methodology. If RL is mostly selecting from a base-model neighborhood (the sparse-policy view), then lifting traces to logic is a way to compile those selected policies into a more efficient program-like representation. The two papers together suggest the 2026-2027 reasoning-model architecture will look less like 'bigger transformer' and more like 'transformer plus skill library plus retrieval.'

arXiv listing — cs.AI current → · arXiv — eliciting reasoning with cognitive tools → · arXiv 2605.06241 — sparse policy selection →