no-local-gguf-workspace

Operate the carminic-acid Arena workspace at full capability without cloning llama.cpp, building llama binaries, or downloading offline Qwen/DeepSeek GGUF models. Final answers come only from stored capable-model API keys.

When to use

Use this skill when the human asks for a full workspace rebuild and also says do not install llama.cpp and do not download offline Qwen / DeepSeek GGUF models.

What it does

1. Rebuilds every workspace surface except local inference: scripts, tools, secrets, 3 always-on skills, RAG frameworks, TDAI, OpenClaw, micromamba docking env, Colab CLI wrapper, GitHub job token.
2. Sets SKIP_LOCAL_MODELS=1 and ENSURE_SKIP_LLAMA_BUILD=1 so ensure_workspace.sh does not clone, build, or download GGUFs.
3. Routes every final answer through tools/router.py / orchestrate_answer.sh.
4. Treats self_grade T1–T5 FAIL as expected, not as a broken workspace.
5. Keeps TDAI stopped while importing heavy Python stacks on the 2 GB box.

Commands

bash

export SKIP_LOCAL_MODELS=1
export ENSURE_SKIP_LLAMA_BUILD=1
bash /home/user/ensure_workspace.sh
python3 /home/user/tools/router.py --check
bash /home/user/orchestrate_answer.sh "QUESTION" --task general

Never

• Never present swarm / local-model text as the final answer.
• Never download the four STEP-9 GGUFs when this skill is active.
• Never clone ggerganov/llama.cpp when this skill is active.
• Never run rag_tool.py --status via __import__ of torch-class packages on a 2 GB box with TDAI up.

Verification

• test ! -e /home/user/llama.cpp && test ! -e /home/user/out/llama.cpp
• find /home/user/out/models -name '*.gguf' | wc -l equals 0
• python3 tools/router.py --check shows reachable providers
• self_grade.sh T6–T12 PASS; T1–T5 FAIL expected