Hugo Railly released `/intent-layer`, the first skill from Crafter Station, designed to improve AI agent performance through better context. The tool works with popular agents like Claude Code and Cursor, aiming to solve inconsistent results when navigating large codebases. It addresses the problem of agents burning tokens exploring dead ends or missing critical config files.

The core issue is that AI agents lack the mental map experienced engineers possess—knowing folder ownership, breaking changes, and real logic paths. Context engineering builds this map by structuring system prompts, inputs, outputs, and tools. Intent Layer focuses on the first piece: creating `AGENTS.md` files at folder boundaries to provide agents with essential context beyond the code itself.

By running the skill, developers can detect existing context files, analyze their codebase structure, and get suggestions for new context nodes. The tool asks what patterns and pitfalls to document, helping agents like the author's own Claude Code find a bug faster by loading 16k tokens instead of 40k. It's open-source, with more context engineering skills planned.