tbench

Automate Terminal-Bench 2.0 benchmarking and failure analysis for Mux agents.

2.0k|126|Updated Sep 17, 2025
One-click install
npx skills add https://github.com/coder/mux --skill tbench-coder
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: tbench
Source: https://github.com/coder/mux/tree/main/.mux/skills/tbench
Command: npx skills add https://github.com/coder/mux --skill tbench-coder

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Terminal-Bench integration for Mux agent benchmarking and failure analysis streamlines evaluating agent performance, diagnosing failures, and iterating improvements in a reproducible way.

Core Features & Use Cases

  • CI/CD benchmarking: run end-to-end Terminal-Bench workloads against Mux agents to compare model performance across configurations.
  • Harbor-based evaluation: leverages Harbor as the evaluation harness for scalable benchmarking in cloud or on-prem environments.
  • Failure analysis workflow: collects per-task results and logs to identify bottlenecks and reliability issues across task sets.

Quick Start

Run a full Terminal-Bench benchmark suite in CI using workflow_dispatch.

Frequently Asked Questions about tbench

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

FAQPage Schema
How do I automate Terminal-Bench benchmarking for agents in a CI/CD pipeline?

Automate Terminal-Bench benchmarking in CI/CD by triggering workflow_dispatch to run end-to-end workloads against agents, compare model performance across configurations, and collect per-task failure analysis results.

What is Terminal-Bench used for in agent performance evaluation?

Terminal-Bench is used for agent performance evaluation to run end-to-end benchmarking workloads, diagnose failures, and iterate agent improvements in a reproducible way across different model configurations and task sets.

Does Harbor work with Terminal-Bench for cloud benchmarking environments?

Yes, Harbor works with Terminal-Bench by acting as the evaluation harness, enabling scalable benchmarking and failure-rate analysis across cloud or on-prem environments for agent workloads.

How do I analyze agent failure rates across different benchmark configurations?

Analyze agent failure rates by collecting per-task results and logs after a Terminal-Bench run, which identifies bottlenecks and reliability issues across various model configurations and task sets.

Best way to compare Mux agent model performance using Terminal-Bench?

Compare Mux agent model performance by running full Terminal-Bench benchmark suites in CI, leveraging Harbor as the evaluation harness to evaluate and contrast results across different configurations.