Show HN: Gauge – Open-Source CLI to recursively improve your docs

Gauge is an open-source CLI for improving how coding agents discover and use a team’s Skills, command-line tools, SDKs, MCP servers, APIs, and documentation. It runs agents through realistic tasks, records their sessions, and surfaces friction such as blocked interactive commands, outdated usage patterns, missed guidance, false claims about limitations, or unclear workflows. Teams can then change code, help text, Skills, or docs and rerun the same tasks to see whether the problem was fixed without introducing new obstacles.

The workflow is designed to run locally with Docker and existing agent subscriptions or API credentials, without requiring a Gauge account. It supports Claude Code, Codex, Pi, and OpenCode; a hosted mode adds testing for documentation and authenticated services. Users define repeatable evaluations with tasks and success criteria, optionally share them as Markdown cases for teammates or CI, and compare results across runs. Gauge handles execution and evidence collection, while a coding agent can analyze transcripts and report findings. The CLI requires Node.js 22.13 or later and Docker; test usage is billed through the user’s agent provider. Saved local evaluations, transcripts, results, and analyses are kept in a private workspace, allowing teams to preserve baselines and validate pre-release versions before publishing.