This resonates with something I've been working on from the opposite direction. Recursive-mode solves context rot for the development process — requirements a

This resonates with something I've been working on from the opposite direction. Recursive-mode solves context rot for the development process — requirements and plans as persistent files instead of chat history. I hit the same problem but with research papers: the knowledge an agent needs to implement a technique from a paper gets lost across sessions. So I built a tool that converts papers into structured markdown files with YAML triggers, every equation with variables defined, and full result tables — designed to sit in the repo alongside the code. The "locked output document" concept maps well here. A skill file is essentially a locked extraction of a paper's implementation-relevant content. Once generated, it doesn't change, and any agent session can consume it without re-reading the source. Interesting to think about combining the two: recursive-mode for the dev workflow, with skill files as the "knowledge layer" that feeds into requirements and planning phases.