tbench

Integrate Mux with Terminal-Bench to automate AI agent benchmarking and failure analysis.

Updated Feb 24, 2026

Search Tags:#ci/cd #benchmarking #llm evaluation #ai agents #performance analysis #terminal-bench

Ready to run

npx skills add https://github.com/neilmovva/mux --skill tbench-neilmovva

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: tbench
Source: https://github.com/neilmovva/mux/tree/main/.mux/skills/tbench
Command: npx skills add https://github.com/neilmovva/mux --skill tbench-neilmovva

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill integrates Mux with Terminal-Bench to automate the benchmarking of AI agents, analyze their performance, and identify areas for improvement.

Core Features & Use Cases

Automated Benchmarking: Run comprehensive benchmark suites against AI agents.
Failure Analysis: Identify tasks where agents underperform compared to others.
Leaderboard Submission: Prepare and submit benchmark results to platforms like Terminal-Bench.
Use Case: Automatically run a suite of tests against a new LLM to gauge its performance on coding tasks and compare it against established models.

Quick Start

Run the full benchmark suite using the make command.

Frequently Asked Questions about tbench

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I benchmark AI agents for coding tasks?▼

Benchmark AI agents by running automated benchmark suites against them to gauge performance on coding tasks and compare results with established models. This process facilitates detailed performance analysis and identifies areas where agents underperform.

How do I analyze failures in LLM evaluation benchmarks?▼

Analyze LLM evaluation failures by identifying specific tasks where AI agents underperform compared to others. This failure analysis highlights weaknesses in agent performance and helps pinpoint areas for improvement.

Can I configure concurrency and timeouts for AI agent benchmarking?▼

Yes, you can configure environment variables for concurrency, timeouts, and task selection during AI agent benchmarking. You can also specify agent-specific arguments for model selection and experiment enablement.

What is the best way to submit AI agent benchmark results to a leaderboard?▼

The best way to submit benchmark results is by preparing and submitting them directly to platforms like Terminal-Bench. This automates the execution of benchmark suites and facilitates the publication of your AI agent performance data.

Do I need to run a full benchmark suite to evaluate a new LLM?▼

No, you do not need to run a full suite if you configure task selection environment variables. You can run targeted tests against a new LLM to gauge its performance and compare it against established models.