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 instructions/metrillm/metrillm/claude-mdgit clone --depth 1 https://github.com/MetriLLM/metrillmWrote 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/metrillm/metrillm/claude-md)<a href="https://agentmods.dev/instructions/metrillm/metrillm/claude-md"><img src="https://agentmods.dev/badge/instructions/metrillm/metrillm/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 | $0.01647 | $0.01647 |
| Opus 5 | $0.00823 | $0.00823 |
| Sonnet 5 | $0.00329 | $0.00329 |
| Haiku 4.5 | $0.00165 | $0.00165 |
Grade A, and why
metrillm 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 5d 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 — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Project
MetriLLM — CLI benchmark tool for local LLM models (Ollama and LM Studio). Measures performance (tok/s, TTFT, memory, CPU load), quality (reasoning, math, coding, instruction following, structured output, multilingual), and computes a hardware fitness verdict.
Commands
npm run build # tsup → dist/index.mjs (ESM, Node20, bundled)
npm run dev # Run from source via tsx
npm run typecheck # tsc --noEmit
npm run test # vitest run
npm run test:watch # vitest watch
npm run test:coverage # vitest with v8 coverage (thresholds enforced)
npm run ci:verify # typecheck + test:coverage + build (full CI check)
npm run test:e2e:smoke # Real Ollama integration test (requires running Ollama)
Run a single test file: npx vitest run tests/scoring.test.ts
Run a single test by name: npx vitest run -t "test name pattern"
Architecture
Entry point: src/index.ts — Commander.js CLI with interactive menu fallback.
Benchmark flow (src/commands/bench.ts):
- Detect hardware (
src/core/hardware.ts) via systeminformation - List/select Ollama models (
src/core/ollama-client.ts) - Run performance benchmark (5 prompts + warmup) →
src/benchmarks/performance.ts - Run quality benchmarks (6 categories) →
src/benchmarks/*.tsagainstsrc/datasets/*.json - Compute scores →
src/scoring/(performance-scorer, quality-scorer, fitness) - Display results →
src/ui/(tables, verdict, colors) - Persist locally →
~/.metrillm/results/ - Optional upload → Supabase via
src/core/uploader.ts
Scoring system:
- Performance score (0-100): speed (50pts) + TTFT (20pts) + memory (30pts), hardware-adaptive tuning by profile (ENTRY/BALANCED/HIGH-END)
- Quality score (0-100): weighted categories — reasoning/coding/instruction=20pts, structured-output/math=15pts, multilingual=10pts. Time penalties for slow responses.
- Global score: 30% performance + 70% quality
- Verdicts: EXCELLENT (>=80) | GOOD (>=60) | MARGINAL (>=40) | NOT RECOMMENDED (<40 or disqualified)
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.
- 5d ago First seen · 109 lines · 1,647 tokens per session scan A 47e6b401c011
metrillm CLAUDE.md is an instructions file published in the GitHub repository MetriLLM/metrillm (7 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 1,647 tokens to every session, about $0.0082 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
oh-my-knowledge AGENTS.md
AGENTS.md instructions for lizhiyao/oh-my-knowledge, covering agents.md - agent 入场清单, 开工先做, 高效交付, 自主 cr 与完成定义 and cr 运行产物隔离.
strix-halo-guide AGENTS.md
AGENTS.md instructions for hogeheer499-commits/strix-halo-guide, covering agents.md, core project lens, documentation style, do-not-invent rules and benchmark claim rules.
openbench AGENTS.md
Instructions for minghinmatthewlam/openbench, covering openbench — agent context, local execution context, what openbench is, execution ownership and product goals (the two things we are building toward).
cai CLAUDE.md
Instructions for cai-layer/cai, covering claude.md, what is cai?, design system, build & run and dev.
foundry-local cpp-download.instructions.md
Use when working on DownloadManager, model caching, model downloads, IsModelCached, or debugging why a model fails to download or loads incorrectly.
The-Ideal-Harness AGENTS.md
AGENTS.md instructions for bharat3645/The-Ideal-Harness, covering agents.md, read this before you write code, 1. enforce below the model, 2. zero overlap and 3. clean-room.