Borrowing it
Nothing to install: this file belongs to JoniMartin27/inferbench. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/JoniMartin27/inferbench/master/CLAUDE.mdgit 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/instructions/jonimartin27/inferbench/claude-md)<a href="https://agentmods.dev/instructions/jonimartin27/inferbench/claude-md"><img src="https://agentmods.dev/badge/instructions/jonimartin27/inferbench/claude-md.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 5.1 | $0.04281 | $0.04281 |
| Opus 5 | $0.02141 | $0.02141 |
| Sonnet 5 | $0.00856 | $0.00856 |
| Haiku 4.5 | $0.00428 | $0.00428 |
Grade A, and why
inferbench CLAUDE.md 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 6d 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 — 149 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Instrucciones para Claude Code en este proyecto
Antes de tocar nada
- Lee
PROJECT_BRIEF.mdpara la visión, fórmulas de compatibilidad y schemas de optimización por motor. README.mdtiene el inventario actual: motores soportados, endpoints, layout de cachés y estado real de cada feature.- Los hitos M1–M9 están todos implementados (ver tabla en README). Esto es mantenimiento y extensión, no greenfield.
Lo que NO debes hacer (load-bearing)
Schemas de optimización son por motor, no uniformes
Cada motor tiene su propio set de flags. llama.cpp usa -ctk, -ctv, --n-cpu-moe, -fa, --n-gpu-layers. vLLM usa otros. APIs cloud (OpenAI, Anthropic, OpenRouter, NVIDIA) sólo admiten parámetros de sampling — no optimización local. Está en PROJECT_BRIEF.md, sección "Schema de optimizaciones POR MOTOR". No intentes unificar.
No simules motores
El prototipo del artefacto inicial usaba datos fake. El proyecto local NO debe simular: si un motor no está disponible o falla, devuelve error claro al frontend. No inventes TTFT, tok/s, ni VRAM. Único mock aceptable: tests unitarios.
No bloquees el event loop
FastAPI es async. Las descargas (binarios GitHub, GGUF de HF), spawn de subprocesos y operaciones Docker se hacen sin bloquear — patrón ya establecido en core/binary_manager.py, core/model_manager.py, core/native_runtime.py. Replica ese patrón.
Secretos
API keys de cloud (OpenAI/Anthropic/etc.) van por keyring (ya es dependencia). No las metas a SQLite ni a archivos de config en plano.
No satures la GPU del display (load-bearing)
vLLM/SGLang/TGI pre-asignan fracción · VRAM_total. En un equipo de UNA GPU que también pinta la pantalla, pedir demasiado ahoga al compositor → cortes de vídeo y cuelgues (pasó de verdad). Por eso:
hardware.safe_gpu_fraction()calcula el tope desde la VRAM libre menos una reserva de display (gpu_display_reserve_gb, por defectomax(2GB, 25%), ajustable con envINFERBENCH_GPU_RESERVE_GB).- Cada motor Docker aplica este tope SIEMPRE en su
build_command/build_environment(vLLM--gpu-memory-utilization, SGLang--mem-fraction-static, TGICUDA_MEMORY_FRACTION), conmin(lo_pedido, lo_seguro). NO dejes que usen su default (~0.9). NO quites el cap. base._start_dockertiene un guard: si no cabe nada de forma segura, lanza error claro en vez de arrancar. Nunca lo bypasees arrancando contenedores GPU a mano sin pasar por el motor.
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.
- 6d ago First seen · 149 lines · 4,281 tokens per session scan A 15f8fbdd914c
inferbench CLAUDE.md is an instructions file published in the GitHub repository JoniMartin27/inferbench (2 stars, last pushed 12d ago), licensed MIT. It adds 4,281 tokens to every session, about $0.0214 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 instructions, from other repositories
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.