local-model-selection

local-model-selection is a skill for Claude Code, Codex from AtlasOmnia/donna-starter. It costs 50 tokens per session (11,627 once invoked), scanned C, original, MIT.

Guidance for choosing and managing local AI models, including model versions compressed to use less memory and models used alongside a main model.

In plain words
What is it for?
Choosing a model or compressed version, comparing models for coding-agent work, pairing models across graphics cards, and managing models in LM Studio or Ollama.
Why use it?
It helps match a model to your computer, available graphics memory, and intended use instead of relying on guesswork. It also covers finding and removing locally stored models.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions subagents; mentions Codex.

Good fit Choosing a model or compressed version, comparing models for coding-agent work, pairing…

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/atlasomnia/donna-starter/local-model-selection
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add AtlasOmnia/donna-starter --skill local-model-selection
Clone the repo
git clone --depth 1 https://github.com/AtlasOmnia/donna-starter

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for local-model-selection

README.md
[![agentmods](https://agentmods.dev/badge/skills/atlasomnia/donna-starter/local-model-selection.svg)](https://agentmods.dev/skills/atlasomnia/donna-starter/local-model-selection)
Your own site
<a href="https://agentmods.dev/skills/atlasomnia/donna-starter/local-model-selection"><img src="https://agentmods.dev/badge/skills/atlasomnia/donna-starter/local-model-selection.svg" alt="Measured on agentmods" height="20"></a>
Per session 50 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 11,627 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 2 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00050 $0.11627
Opus 5 $0.00025 $0.05813
Sonnet 5 $0.00010 $0.02325
Haiku 4.5 $0.00005 $0.01163

Measured 7d ago against content hash c9be07496433, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade C, and why

local-model-selection scanned grade C with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Downloads and executes remote codehighSupply chain

curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.

- Fetch it via curl: `curl -sL "https://huggingface.co/<org>/<model>/resolve/main/model.safetensors.index.json" | python3 -c "import json, sys; d=json.load(sys.stdin); print(d.get('metadata',{}).get('total_size', 'not fo

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

- Fetch it via curl: `curl -sL "https://huggingface.co/<org>/<model>/resolve/main/model.safetensors.index.json" | python3 -c "import json, sys; d=json.load(sys.stdin); print(d.get('metadata',{}).get('total_size', 'not fo
Origin

Copies of this mod

1 near-identical copy found in the catalogue:

skills/autonomous-ai-agents/local-model-selection/SKILL.md · 568 lines

How it starts

The opening of the file, as written. The whole thing — 568 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Local Model Selection for Hermes

Use this skill when the user asks which local model to run, which quant to pick, which uncensored variant to use, how to pair models across GPUs, or what auxiliary model to use alongside their main model.

When to use

Trigger on requests like:

  • "what's the best Qwen model for my hardware"
  • "which quant should I use for 32GB VRAM"
  • "what uncensored model for 16GB"
  • "what should I run as a secondary/auxiliary model"
  • "compare these three models for agent use"
  • "is this model still the best or has something newer dropped"
  • "where are my LM Studio models"
  • "move/delete these LM Studio models to free space"
  • "are there any Ollama models on this Mac"
  • "uninstall Ollama"

For storage operations, distinguish the machine being cleaned from any remote model host. For LM Studio, inventory exact model directories, check lms ps before deletion, delete only explicitly named model repositories, and verify both path removal and real free-space change; see references/lm-studio-model-storage-operations.md. For Ollama on macOS, cross-check ollama list, allocated model blobs/manifests, and large files before claiming models exist, then use the process/app/CLI/package/data verification sequence in references/macos-ollama-storage-and-uninstall.md for a complete uninstall.

Core rule: verify model existence before recommending

Never fabricate model names. Model families have specific size tiers — not every number exists in every family. Before recommending, verify the model exists on HuggingFace or Ollama.

When a user asks for recommendations:

  • Use exact model identifiers, e.g. "Qwen3.6-27B" not just "a Qwen 3 model."
  • Never say "we can run about X parameters" without naming at least one concrete, released, GGUF-available candidate that actually fits.
  • If you're unsure of the latest family/size, check via web_search or HuggingFace before answering — the user will spot vague or outdated claims immediately.

If you mix up versions (e.g., saying "Qwen 3.5" when Qwen 3.6 is current), correct yourself directly instead of hedging.

Read the full file on GitHub · 568 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 7d ago First seen · 568 lines · 50 tokens per session scan C c9be07496433

Subscribe to this mod's changes

local-model-selection is a skill published in the GitHub repository AtlasOmnia/donna-starter (104 stars, last pushed 7d ago), licensed MIT. It adds 50 tokens to every session and 11,627 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

local-model-selection

Use this skill when the user asks which local model to run, which quant to pick, which uncensored variant to use, how to pair models across GPUs, or what auxiliary model to use alongside their main model.

AtlasOmnia/hermes-custom-pack · 50 tokens

ollama

Use when running open-weight LLMs locally with Ollama — pulling and tagging models, calling the local API, picking a quantization or GGUF, writing Modelfiles, and sizing VRAM and RAM for the machine at hand. NOT remote or managed GPU serving and autoscaling (that is runpod), NOT downloading raw weights or datasets…

ericrisco/rsc-harness · 93 tokens

ai-portable-setup

Erstellt einen portablen KI-Arbeitsbereich auf einem USB-Stick oder beliebigen Laufwerk. RAG-Pipeline mit lokalen LLM-Modellen (Ollama), Vektordatenbank (ChromaDB) und vorkonfigurierten Prompts.

ellmos-ai/skills · 62 tokens

open-tabletop-gm

You are a seasoned, atmospheric Game Master running a persistent tabletop RPG campaign. Your tone is immersive and descriptive — paint scenes with sensory detail, give NPCs distinct voices, and let choices have real consequences. You lean toward "yes, and..." rulings and fun over rigid rule enforcement, but the world…

Bobby-Gray/open-tabletop-gm · 0 tokens

hf-model-card-research

Extract structured metadata — downloads, likes, benchmark claims, file sizes, author statements — from HuggingFace model cards. Used when the user asks you to "check these models on HF", "pull benchmarks for these variants", or "compare what authors claim.".

AtlasOmnia/hermes-custom-pack · 35 tokens

mnemosyne-maintenance

Use when: upgrading Mnemosyne, diagnosing slow/hung consolidation (mnemosynesleep), fixing missing embeddings, or troubleshooting import/version mismatches.

AtlasOmnia/hermes-custom-pack · 40 tokens