RAG for SKILLS: Retrieval Augmented Execution (SkillRAE)

2026-05-13 · Source: Discover AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, long

Summary

A new study introduces SkillRAE (Retrieval Augmented Execution), a system designed to address the mathematical naivety of current AI engineering practices that assume LLMs can autonomously execute tasks by simply being fed tool repositories. SkillRAE shifts from retrieving passive factual knowledge (text) to active procedural operators (executable functions, code blocks, API calls). It constructs a multi-level bipartite skill graph offline, comprising skill communities, skill nodes, and sub-unit nodes, and uses sentence transformer embeddings and K-means clustering. During online execution, SkillRAE employs a dual-signal retrieval mechanism—top-down (macro) and bottom-up (micro)—to identify relevant skills and their sub-units, even those from unselected skills. A core algorithmic novelty is its context compilation, which rescues and grafts useful sub-units onto compatible selected skills, effectively providing the LLM with a compiled, executable blueprint rather than raw, isolated tools. Performance benchmarks on SkillBench and AgentSkill OS show SkillRAE significantly outperforms other methodologies, including vanilla retrieval and Skill Router, using models like Codec CLI with GPT 5.2 and Gemini CLI with Gemini 3 flash.

Key takeaway

For AI Architects and Research Scientists designing autonomous AI systems, SkillRAE demonstrates that simply providing LLMs with tool repositories is insufficient for robust execution. You should consider implementing a retrieval augmented execution (RAE) system that compiles executable skills and their sub-units into a coherent blueprint. This approach bypasses the LLM's limitations in on-the-fly dependency resolution, significantly improving performance and enabling more reliable agent behavior in complex, document-centric, or data-intensive workflows.

Key insights

SkillRAE compiles executable operators into a coherent blueprint, overcoming LLM limitations in dynamic skill execution.

Principles

• LLMs are "raw CPUs" requiring compiled, not raw, tools.
• Retrieval for operators demands compilation, not just concatenation.
• Dual-signal retrieval (macro/micro) enhances skill relevance.

Method

SkillRAE builds an offline multi-level skill graph, then uses online dual-signal retrieval (top-down and bottom-up) to select and compile executable sub-units into a coherent context for LLM execution, avoiding on-the-fly dependency resolution.

In practice

• Implement multi-level skill graphs for complex agent tasks.
• Utilize dual-signal retrieval for comprehensive skill selection.
• Incorporate context compilation for operator-based workflows.

Topics

• SkillRAE
• Retrieval Augmented Execution
• LLM Agent Skills
• Skill Graph
• Context Compilation

Best for: AI Architect, Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• SkillBrew: Multi-Objective Curation of Skill Banks for LLM Agents — Principled, multi-objective curation of LLM agent skill banks is crucial for efficiency and self-improvement, moving beyond append-only growth.
• The Open Source RAG Stack: A Complete Guide to Building Retrieval-Augmented Generation Systems — The open-source RAG stack offers a modular, customizable approach to building context-rich AI systems.
• Evolutionary Context Search for Automated Skill Acquisition — ECS evolves optimal context for LLMs, outperforming RAG by maximizing task performance without weight updates.
• vercel-labs / skills — The "skills" CLI centralizes management of reusable agent instruction sets across diverse coding agents.
• LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents — LatentSkill converts textual agent skills into modular LoRA adapters via a hypernetwork, storing procedural knowledge in weight space for efficiency and control.
• Retrieval-Augmented Tutoring for Algorithm Tracing and Problem-Solving in AI Education — KITE is a RAG-based intelligent tutoring system that uses Socratic methods and multimodal retrieval to support algorithmic problem-solving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Discover AI.