A literate-programming grammar with a first-class name graph gives coding agents IDE-style symbol search without an IDE

arXiv 2608.24644·low signalXBlueskyLinkedInCopy link

arXiv 2608.24644 (2026-08-25) describes an environment for writing executable program essays, with a parser that treats names as first-class objects and an internal name graph relating prose, names and executable artifacts, plus bindings to existing languages and test tooling. The stated goals are to cut what an LLM needs in context by co-locating code with the prose that explains it, and to give agents symbol-aware search and usage tracking without a language server. Three languages are supported in the implementation with example programs; single-author, no repo link, so treat it as a design worth stealing rather than a tool to adopt.