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.
npx agentmods add skills/av/harbor/run-llmsnpx skills add av/harbor --skill run-llmsgit clone --depth 1 https://github.com/av/harborWrote 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.
[](https://agentmods.dev/skills/av/harbor/run-llms)<a href="https://agentmods.dev/skills/av/harbor/run-llms"><img src="https://agentmods.dev/badge/skills/av/harbor/run-llms.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00114 | $0.10701 |
| Opus 5 | $0.00057 | $0.05351 |
| Sonnet 5 | $0.00023 | $0.02140 |
| Haiku 4.5 | $0.00011 | $0.01070 |
Grade F, and why
run-llms scanned grade F with 4 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 4d 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
4. Restart: sudo systemctl restart docker && harbor down && harbor up 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.
curl https://av.codes/get-harbor.sh | bash Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
rm -rf services/openterminal/data Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl https://av.codes/get-harbor.sh | bash How it starts
The opening of the file, as written. The whole thing — 1,322 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Run LLMs Locally with Harbor
Harbor is a containerized LLM toolkit. This skill enables autonomous setup, configuration, troubleshooting, and operation of local LLM infrastructure.
Agent Decision Trees
Use these decision trees to determine what action to take for common user requests.
User wants to run an LLM
1. Is Harbor installed?
→ NO: Install Harbor (see Initial Setup)
→ YES: Continue
2. Is Docker running?
→ Run: docker info
→ FAIL: Start Docker daemon, check installation
→ OK: Continue
3. Does the user have a specific model in mind?
→ YES: Determine format (Ollama tag, GGUF, HF safetensors)
→ Ollama tag (e.g. qwen3:4b): harbor pull <model> && harbor up
→ GGUF from HuggingFace: harbor pull <org/repo> && harbor up llamacpp
→ Safetensors/HF model: harbor vllm model <user/repo> && harbor up vllm
→ NO: Recommend a small default: harbor pull qwen3:4b && harbor up
4. Verify: harbor ps → confirm services healthy
5. Open UI: harbor open
User has GPU issues
1. Check NVIDIA drivers: nvidia-smi
→ FAIL: User needs to install NVIDIA drivers
→ OK: Continue
2. Check Container Toolkit: docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi
→ FAIL: Install NVIDIA Container Toolkit, restart Docker
→ OK: Continue
3. Check service logs: harbor logs <service> # ⚠️ TAILS INDEFINITELY! (Agents: use `docker logs harbor.<service>` instead)
→ Look for: "CUDA error", "out of memory", "no GPU"
→ OOM: See "Model won't load / OOM" troubleshooting
→ No GPU detected: Check /etc/docker/daemon.json for nvidia runtime
4. Restart: sudo systemctl restart docker && harbor down && harbor up
User wants web search in chat
1. Start SearXNG: harbor up searxng
→ SearXNG auto-wires to Open WebUI when both run together
2. If WebUI was already running: harbor restart webui
3. Verify: harbor ps | grep searxng
4. Open UI: harbor open → Web search is now available in chat
User wants to change the model
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.
- 4d ago First seen · 1,322 lines · 114 tokens per session scan F b7bc007b97e4
run-llms is a skill published in the GitHub repository av/harbor (3,202 stars, last pushed 4d ago), licensed Apache-2.0. It adds 114 tokens to every session and 10,701 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it F with 4 findings (asks for root, downloads and executes remote code, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
gws-modelarmor-sanitize-prompt
Google Model Armor: Sanitize a user prompt through a Model Armor template.
knowledge-base
使用 Yuxi 知识库进行检索、打开文档、文档内定位和查看思维导图。当用户需要基于已配置知识库回答问题、核验资料或引用文档内容时使用此技能。.
markitdown
Convert heterogeneous documents and selected URIs to Markdown with Microsoft MarkItDown for text analysis, search, and LLM/RAG ingestion. Covers safe local conversion, streams, Office/PDF/data formats, batch workflows, plugins, vision OCR, Azure extraction, and the official MCP server.
rag-knowledge
Work with the RAG knowledge base — ingest documents, run semantic search, manage collections, or add a sync source/connector (Google Drive, S3). Use when populating or debugging the knowledge base, tuning retrieval, or adding a new document source. This project uses {{ cookiecutter.vectorstore }} + {{…
claude-api
Build, debug, and optimize Claude API / Anthropic SDK apps. Apps built with this skill should include prompt caching. Also handles migrating existing Claude API code between Claude model versions (4.5 → 4.6, 4.6 → 4.7, retired-model replacements). TRIGGER when: code imports anthropic/@anthropic-ai/sdk; user asks for…
ingest
Turn external URLs and documents into LLM-ready content — load + cache web pages (HQCC compression) and OCR PDFs/images to Markdown. Use whenever the user gives a URL, asks you to read a webpage, or attaches a PDF/scan that needs to be parsed before reasoning. Executes via the cloud load, cloud search, and cloud parse…