PODSHL: bug reports that read the GPU, driver and runtime version off the reporter's machine, with their consent

[dx111ge]

llama.cpp gets a lot of "doesn't work on my GPU" reports (500+ issues in the last month), and most of them turn on the same few facts: which GPU, how much VRAM, which driver, which compute capability, which build. Those are also the facts reporters get wrong most often, because they answer from memory.

I've been building an open-source client that reads them on the user's machine instead. A project lists the facts it needs in one YAML file in its repo; the user sees each value before anything leaves the machine, and the report marks every fact as measured or typed. What can be read is a fixed list compiled into the client (no arbitrary files, no commands a repo can choose), and for GPUs today that is nvidia-smi --query-gpu=... (name, driver, VRAM, compute capability, temperature, power; UUID/PCI/VBIOS fields are refused), lspci, system_profiler on macOS, OS and kernel version, and <program> --version for a named program. It also works the other way round: a project can publish known fixes as small Markdown files, and the client checks the user's machine against them before anyone opens an issue.

The worked example is a project of mine that talks to Ollama / any OpenAI-compatible endpoint: "the chat never answers" turns into three readings (endpoint variable, program version, Ollama version or container tag) and one of five published answers, or a report with those facts attached.

What it can't do yet, and where I'd like your input:

• the backend a llama.cpp binary was built with (CUDA/Vulkan/Metal/ROCm) and its build flags: --version output is reduced to the version number today
• AMD/ROCm and Intel GPUs: there is no rocm-smi or equivalent in the list yet
• anything from a running llama-server

If you triage issues here: which facts do you ask for most often? That tells me what to add first, and I'm happy to write the YAML for llama.cpp so you can try it without touching the repo.

AGPL-3.0, spec Apache-2.0: https://github.com/dx111ge/podshl