google/benchmark is a C++ library for measuring the performance of small code sections, much like tests measure correctness. Developers use it to run repeatable microbenchmarks and compare execution results.
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/google/benchmark/agents-mdgit clone --depth 1 https://github.com/google/benchmarkWrote 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/google/benchmark/agents-md)<a href="https://agentmods.dev/instructions/google/benchmark/agents-md"><img src="https://agentmods.dev/badge/instructions/google/benchmark/agents-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 | $0.00410 | $0.00410 |
| Opus 5 | $0.00205 | $0.00205 |
| Sonnet 5 | $0.00082 | $0.00082 |
| Haiku 4.5 | $0.00041 | $0.00041 |
Grade A, and why
benchmark AGENTS.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 yesterday.
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.
What it actually says
AI usage
[!NOTE] The modern (as of 2026-06) AI is a misnomer,
- it is not conscious / not a consciousness,
- it is not sentient,
- it is not intelligent,
- it does not think,
- it does not understand the code,
- it is merely a next-token guesser,
... therefore it is merely an (hyper-) advanced IDE.
[!NOTE] A contribution is any externally-observable interaction with a project.
[!CAUTION] Failure to follow the following rules may result in repercussions, possibly without a prior warning.
Rules:
- It is acceptable to use AI when producing contributions.
[!WARNING] Any and all AI usage MUST be fully and explicitly disclosed in every contribution.
- All contributions shall be done by conscious, sentient beings. Fully autonomous contributions by bots are prohibited1.
[!WARNING] Dear contributor, the AI is your tool, and its output is for your consumption. It is your responsibility to consume said output, interpret it, and then produce the contribution itself. DO NOT just query it and post the output, especially so for all non-code contributions!
- The contributor (conscious, sentient being) solely bears the whole responsibility for the contribution, they must understand the problem, and the solution, and be able to constructively argue about it. "well, AI said so, therefore it is" is not an acceptable approach.
Footnotes
-
Unless explicitly allowed by maintainers on case-by-case basis before the contribution is submitted, in which case the bot owner (conscious, sentient being) is recognized as the de facto contributor. E.g. https://github.com/apps/dependabot is allowed. ↩
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.
- yesterday First seen · 45 lines · 410 tokens per session scan A b6065ae9c491
benchmark AGENTS.md is an instructions file published in the GitHub repository google/benchmark (10,385 stars, last pushed 8d ago), licensed Apache-2.0. It adds 410 tokens to every session, about $0.0020 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 2026-09-03.
Other instructions, from other repositories
aws-bench AGENTS.md
AGENTS.md instructions for aws-bench/aws-bench, covering agents.md, setup, commands, project structure and code style.
open-dictionary AGENTS.md
Instructions for ahpxex/open-dictionary, covering open dictionary rewrite charter, product framing, core workflow, technical framework and 1. raw ingestion layer.
ab900 CLAUDE.md
Instructions for timothywarner-org/ab900, covering claude.md, repository purpose, architecture, commands and 2. non-ascii punctuation -- must return zero.
canvas-lms-mcp AGENTS.md
AGENTS.md instructions for bruchris/canvas-lms-mcp, covering agents.md — canvas lms mcp server, quick start, run with npx (no install needed), or install globally and architecture.
obsidian-university-workflow CLAUDE.md
Instructions for ABO896/obsidian-university-workflow, covering obsidian university workflow, project structure, rules for working on this project, templater api — always read the docs first and config alignment.
agentic-ai-engineering-course AGENTS.md
AGENTS.md instructions for towardsai/agentic-ai-engineering-course, covering the what and course map (lessons ↔ code projects).