AI

SPEM: Structure-Aware Procedural Memory for Efficient Skill Use in LLM Agents

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 21/10/2026, 04:03:25 UTC.

Briefings

SPEM is a new structure-aware procedural memory framework for LLM agents that improves skill reuse by maintaining separate memory types for atomic skills and reusable skill compositions. It achieves 87.1% success on ALFWorld and 38.38% reward on SkillsBench while reducing token and runtime costs compared to strong baselines.

Key points

• SPEM maintains two memory types: Atomic Skill Memory for individual skills and Skill-DAG Memory for reusable compositions derived from successful executions.
• The framework uses Skill-GNN for state- and structure-aware skill selection, reusing high-confidence Skill-DAG templates when available for new tasks.

Context

LLM-based agents increasingly rely on reusable skills, but scaling skill libraries makes efficient use difficult due to high token and runtime overhead. Existing methods treat skills as independent text entries, ignoring execution order and state transitions. SPEM addresses this by proposing a lightweight structure-aware procedural memory framework.

Limitations

Evidence is from an arXiv preprint (arXiv:2610.26995v1) with no peer review. Results are limited to ALFWorld and SkillsBench benchmarks. Token and runtime cost reductions are relative to strong baselines but absolute values are not provided. No causal claims about long-term agent behavior are supported.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 21/10/2026, 04:24:20 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeibc6gfe6enwrrzde7kedxjtw3zr62oxvuyowx5ybat3j4nfk6nu7m