# Gauge open-sources CLI to test how coding agents use your tools and docs

published: 2026-11-05T07:05:36.560Z
topic: infrastructure
source: https://github.com/gauge-sh/gauge-cli
url: https://aiagentnews.io/stories/gauge-open-sources-cli-to-test-how-coding-agents-use-your-tools-and-docs

Gauge has released an open-source CLI that runs coding agents against a product and records where they get stuck. It is aimed at improving Skills, CLIs, SDKs, MCP servers, APIs and docs so that agents can find and use them.

Gauge is installed from npm as @withgauge/cli and is built to measure how easily coding agents discover and use Skills, CLIs, SDKs, MCP servers, APIs and documentation. It runs real agents against a product, captures the points where they stall, and lets a coding agent test proposed fixes before anything is published. Findings can include interactive commands that block an agent, use of deprecated commands, and Skills that go unused.

Local runs use Docker on the user's machine together with existing Claude Code, Codex, Pi or OpenCode logins or API keys, and no Gauge account is required. The stated requirements are Node.js 22.13 or later, npm, and Docker running Linux containers. Local mode supports Skills, SDKs, CLIs and files, while docs, authenticated MCP servers and authenticated APIs need the hosted option.

The workflow runs in a loop of defining tasks, testing the current product, analyzing friction, changing the product and retesting. Gauge also has an optional hosted mode with managed compute and shared results. The CLI collects anonymous command usage by default, and the README says this can be turned off with gauge telemetry disable. The project is licensed under MIT.

## Why it matters if you build agents

Developers who ship tools for agents can check whether an agent completes a task with their Skill, CLI or docs before release, instead of learning about the problem from users. Local runs bill agent usage to the developer's own provider or subscription. Teams that do not want command usage data sent should disable telemetry before running it in CI.