aide-backend-autoselect SKILL.md skillA
aide-backend-autoselect from anonymousnomad/aide-sovereign-workbench is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add anonymousnomad/aide-sovereign-workbench/aide-backend-autoselect@git:20260826.8da8b03Pins git:20260826.8da8b03 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- anonymousnomad/aide-sovereign-workbench
- Path
skills/packs/aide-backend-autoselect/SKILL.md- Versions recorded
- 1
- Latest version
- git:20260826.8da8b03, committed Aug 26, 2026
- Size
- 5,573 bytes
- Repository stars
- 53
- License
- Apache-2.0
Backend auto-detection and selection SOP for llama-server engines in AIDE — probes candidate binaries via --list-devices, caches capabilities, selects per-model backend/offload (Vulkan/CUDA/CPU) from operator profile override > benchmark verdict > heuristic, ships multi-binary awareness so every download gets optimal tok/s out of the box. Use whenever wiring engine spawn logic, adding GPU offload controls, diagnosing "model ignores -ngl" or slow generation, deciding which llama.cpp builds ship…
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Backend Auto-Select — Right Engine, Every Machine Born 2026-08-25: operator reported Vulkan gives better coherence/throughput than the configured default. Investigation found the DEFAULT llama-cpp build was CPU-only (empty --list-devices) while -ngl flags were silently ignored — every AIDE model had been running pure CPU unnoticed. Adjacent builds existed: llama-cpp-cuda (no visible devices on this driver) and llama-cpp-vulkan (GTX 1060 @ 6245 MiB detected). This skill makes that discovery automatic. ## Research base (verified 2026-08-25) 1. **llama.cpp issue #19817 — GTX 1060 exact match**: Vulkan slower at prompt processing, NOTICEABLY FASTER token-generation vs CUDA on Pascal (granite-1B tg128: 90.63 vs 61.65 t/s = +47%; falcon-7B +11%). Pascal CUDA k-quants historically weak (PR #1930 era; LLAMA_CUDA_KQUANTS_ITER tweaks; one 1060 tester's best TG was zero-offload CPU). 2. **NVIDIA forum cross-vendor bench (Mar 2026)**: tg roughly parity Vulkan/CUDA on modern cards, pp favors CUDA ~15%; GGML_VK_PREFER_HOST_MEMORY=1 adds 10-15% Vulkan tg. Discussion #23109: qwen3-0.6B GT1030-class Vulkan tg128 …
Read the whole file at its exact version.
How to install
mdr add anonymousnomad/aide-sovereign-workbench/aide-backend-autoselect@sha256:7a5c0e1dd55d9682The line at the top of this page pins the label git:20260826.8da8b03, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_klks2rvpslrcycco)
4 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 1 version of this file, committed on Aug 26, 2026.
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (5573 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
anonymousnomad/aide-sovereign-workbench · 53 stars · license Apache-2.0 · pushed 2026-10-26 · branch covert-production
API
GET https://markdownregistry.com/api/v1/artifacts/art_klks2rvpslrcycco GET https://markdownregistry.com/api/v1/resolve?ref=anonymousnomad/aide-sovereign-workbench/aide-backend-autoselect GET https://markdownregistry.com/api/v1/blob/7a5c0e1dd55d96820b78371b31716d1c7c3e3eec90d11112c68930d6be382a62
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.