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 skills add JoniMartin27/inferbench --skill run-inferbench-backendgit clone --depth 1 https://github.com/JoniMartin27/inferbenchWrote 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/jonimartin27/inferbench/run-inferbench-backend)<a href="https://agentmods.dev/skills/jonimartin27/inferbench/run-inferbench-backend"><img src="https://agentmods.dev/badge/skills/jonimartin27/inferbench/run-inferbench-backend/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/jonimartin27/inferbench/run-inferbench-backend"><img src="https://agentmods.dev/badge/skills/jonimartin27/inferbench/run-inferbench-backend.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00090 | $0.01802 |
| Opus 5 | $0.00045 | $0.00901 |
| Sonnet 5 | $0.00018 | $0.00360 |
| Haiku 4.5 | $0.00009 | $0.00180 |
Grade A, and why
run-inferbench-backend scanned grade A with 0 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 10d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Run InferBench backend + engines
InferBench's backend is a Python 3.11 FastAPI app on :7777 that orchestrates
local inference engines (downloads binaries/GGUFs, starts engines natively or via
Docker, benchmarks them, persists to SQLite). The Electron frontend is just a REST/SSE
client of it. This skill drives the backend programmatically via
driver.py, which manages the uvicorn lifecycle, smoke-tests the key
endpoints, and runs a real benchmark over the production HTTP+SSE path — asserting
real tokens, no false positives.
All paths below are relative to the inferbench/ project root. Environment verified:
Windows 11, Docker Desktop 29.4.2, NVIDIA RTX 3070.
Prerequisites
The backend needs its uv venv (Python 3.11). If backend/.venv/ is missing:
cd backend && uv venv --python 3.11 && uv pip install -e ".[dev]"
The driver imports httpx, so it must run with the backend venv's Python
(backend/.venv/Scripts/python.exe on Windows). Engines also need, on demand:
Docker Desktop running + an NVIDIA GPU for vllm/sglang/tgi; nothing extra for
llamacpp (downloads the official binary) or ollama (needs Ollama installed).
Run (agent path) — this is the primary path
Smoke test (spawns uvicorn, checks health/hardware/engines/models, shuts down):
backend/.venv/Scripts/python.exe ".claude/skills/run-inferbench-backend/driver.py"
Full drive — smoke plus a real end-to-end benchmark through POST /api/benchmark/run
- the SSE stream (auto-bootstrap → engine start → inference → result).
llamacpp+ the smallest catalog model is the fastest honest check (~370 tps on the RTX 3070):
backend/.venv/Scripts/python.exe ".claude/skills/run-inferbench-backend/driver.py" --benchmark llamacpp:smollm2-360m
Other verified engines (each starts a real container/daemon and runs a model — slower,
need Docker+GPU / Ollama): vllm:qwen2.5-0.5b, sglang:qwen2.5-0.5b, tgi:qwen2.5-0.5b,
ollama:smollm2-360m.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 10d ago First seen · 129 lines · 90 tokens per session scan A e7ee4017acd3
run-inferbench-backend is a skill published in the GitHub repository JoniMartin27/inferbench (2 stars, last pushed 16d ago), licensed MIT. It adds 90 tokens to every session and 1,802 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
groq-inference
Ultra-fast LLM inference on custom LPU hardware. OpenAI-compatible API at api.groq.com. Lowest latency in the industry (500-1000+ tok/s). Supports chat completions, vision, audio (Whisper STT + TTS), tool calling, JSON mode, and streaming. Free tier available. Inference only — no training.
apple-calendar
Read macOS Calendar events via the icalBuddy CLI and create events via AppleScript (osascript). Use to check the user's calendar, agenda, upcoming events, or add an event on macOS.
apple-reminders
Manage Apple Reminders via the remindctl CLI on macOS — list, add, complete, delete, manage lists.
github
Drive GitHub via the official gh CLI — repos, issues, pull requests, releases, gists, Actions runs, and raw REST through gh api. Use when the user asks to inspect or manage GitHub.
obsidian
Read, search, and create Markdown notes inside an Obsidian vault on disk.
ffmpeg
Process audio and video with the ffmpeg / ffprobe CLIs — convert, trim, extract audio, resize, change format, make GIFs, inspect media. Use for any audio/video transformation.