mulch

[Adoption assessment] There is evidence of attention, but not of proven usefulness yet. The strongest source is the official GitHub repository showing roughly 320 stars and 40 forks, which is a heat signal: people are noticing it and some want to follow or experiment. But there are no hands-on reviews, tutorials, long-form implementation writeups, production case studies, or substantial issue discussions in the provided evidence, so there is little usability proof. Based on the current sources, mulch looks like an early open-source idea with visible interest, not a battle-tested standard.

[What it actually does] mulch proposes storing “knowledge for coding agents” as structured files that live in Git with the codebase, so any agent can reuse team conventions, domain rules, architecture preferences, and historical decisions. It is not a general chatbot and not a full RAG retrieval stack. A more accurate analogy is a project-local memory layer for coding agents, or prompt/rule assets versioned with code. Its intended output is not the final software artifact itself, but a maintainable layer of reusable agent context that reduces repeated explanation.