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/przeprogramowani/10x-bench/claude-mdgit clone --depth 1 https://github.com/przeprogramowani/10x-benchWrote 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/przeprogramowani/10x-bench/claude-md)<a href="https://agentmods.dev/instructions/przeprogramowani/10x-bench/claude-md"><img src="https://agentmods.dev/badge/instructions/przeprogramowani/10x-bench/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 | $0.02013 | $0.02013 |
| Opus 5 | $0.01007 | $0.01007 |
| Sonnet 5 | $0.00403 | $0.00403 |
| Haiku 4.5 | $0.00201 | $0.00201 |
Grade A, and why
10x-bench 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 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.
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 — 208 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 Overview
10xBench is a comprehensive benchmark comparing how different large language models handle "vibe coding"—creating a fully functional website for Przeprogramowani.pl in a single attempt without iterative refinement.
Each of four models (Claude Opus 4.6, GPT-5.3-Codex, Kimi K2.5, GLM-4.7) receives the same prompt and must generate a complete website implementation. Results are systematically evaluated against defined criteria and presented in an interactive dashboard.
Quick Start Commands
Development & Building
# Process results CSV files and generate results.json
npm run process-results
# Start development server (Astro) with results processing
npm run dev
# Build production site
npm run build
# Development from website directory (for Astro only)
cd website && npm run dev
cd website && npm run build
Evaluation
# Run the evaluation skill to assess an implementation
/10x-score-attempts against <attempt-directory>
# Example:
/10x-score-attempts against claude opus attempt 1
Project Structure
Core Directories
-
./scripts/— Data processing utilitiesprocess-results.ts— TypeScript script that parses CSV evaluation files from each attempt and generatesresults.json
-
./website/— Astro-based results dashboardsrc/pages/index.astro— Main results overview pagesrc/pages/benchmark.astro— Benchmark prompt and criteria pagesrc/components/ResultCard.tsx— Summary cards showing attempt scoressrc/components/ResultsTable.tsx— Interactive comparison table with sticky headerssrc/components/ModelAveragesCard.tsx— Model family average summariessrc/data/results.json— Generated data file (created byprocess-results.ts)
-
./eval-attempts/— Model implementations (one-shot attempts)claude-opus-attempt-{1,2,3}/— Claude Opus 4.6 implementationsgpt-codex-attempt-{1,2,3}/— GPT-5.3-Codex implementationskimi-k2.5-attempt-{1,2,3}/— Kimi K2.5 implementationsglm-4.7-attempt-{1,2,3}/— GLM-4.7 implementations- Each contains
eval-result.csvwith criterion-by-criterion scores
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 · 208 lines · 2,013 tokens per session scan A b196a6d1d7cd
10x-bench CLAUDE.md is an instructions file published in the GitHub repository przeprogramowani/10x-bench (11 stars, last pushed 16d ago), licensed MIT. It adds 2,013 tokens to every session, about $0.0101 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 2026-08-30.
Other instructions, from other repositories
hiveshare CLAUDE.md
Instructions for KB-perByte/hiveshare, covering hiveshare — claude.md, repo layout, build & run, key env vars (server) and naming: hive vs memory.
llm-safe-haven CLAUDE.md
Instructions for pleasedodisturb/llm-safe-haven, covering llm safe haven, what this is, project structure, tdd — non-negotiable (adopted 2026-08-17) and the contract.
kleosrules AGENTS.md
Instructions for kleosr/kleosrules, a project described as: Cursor harness pack: user rules, skills, Bash hooks, local HANDOFF memory. macOS, Linux, Windows (WSL).
finding-unknowns-skills CLAUDE.md
Claude Code instructions for Neeeophytee/finding-unknowns-skills, covering repo notes, what ships, and the gate that decides, version lives in three manifests — move them together, files that must be edited together and shipped guidance is not this file.
coding-agent-safety-gate AGENTS.md
Instructions for ASER-ho/coding-agent-safety-gate, covering agents / 代理规则, 仓库类型 / repository type and ai 代理规则 / rules for ai coding agents.
data-scientist copilot-instructions.md
Instructions for realnghon/data-scientist, covering github copilot — 项目指令：data scientist 工作流 and 共享核心 — 回答前请阅读.