[paulhothersall]

Title

Issue: Resolves #909

Summary

Docs only, no engine code: an eight-line contributing section in AGENTS.md and docs/TEST_REQUESTS.md, a five-row table of hardware wanted that points at docs/COMMUNITY_BENCHMARKS.md. As offered in #909 and accepted there; the wording is yours to edit or cut.

What changed

• AGENTS.md: new section "Contributing a change or report" (search first, use the issue forms, one change per PR, opt-in with the default path unchanged, build every backend a file touches, change a default only where measured faster, reports from other hardware follow the benchmark format). The rules come from your review comments on Multi-GPU: a --layer-split stage holds only its own layers' weights, and every card's leftover VRAM spills into a helper expert tier (4-way IQ3_S: 9,269 -> 14,172 expert slots, prefill chunk 2,048 -> 4,096) #390, Heterogeneous multi-GPU roles: whole model on one GPU, the MTP draft head on the other #492, Prefill chunk ring #583, refill: improve multi-GPU processing speed ~27% PP improvement on 4× RTX 3060 #598 and prefill: a layer split's idle card streams a share of a one-chunk prompt's experts #663.
• docs/TEST_REQUESTS.md (new): unequal NVIDIA pair, 3-4 NVIDIA GPUs, Pascal (sm_61), Linux host with 32 GB RAM or less, long soaks. Each row is taken from what you wrote on earlier threads, plus Pascal, which I can run.

Extra Notes

I drafted this with an AI assistant (Claude Code) and checked every issue number against the repo myself.

[Niko1221]

Thanks @paulhothersall. The Contributing section in AGENTS.md and the test-requests list are in 0.1.41 with your authorship kept. Closing as ported.