Evan from Gauge announces the open-source release of their CLI to improve documentation for coding agents. The tool features a local execution backend and a hillclimbing loop that iteratively refines docs based on agent feedback. The CLI supports optimizing agent skills, CLI interfaces, and other agent-facing surfaces.
Its design centers on VMs for agent execution and a network gateway that terminates TLS and transparently swaps doc variants. The gateway also handles credential replacement to prevent key exfiltration. The team used the loop internally to fix issues like mistaken pricing and inaccurate product limitations.
For example, they helped Clerk shift agents from legacy 'keyless' mode to 'clerk init', resulting in a 3x increase in usage post-deploy. The system leverages closed-loop optimization, allowing the same agent to be simulated repeatedly in a high-fidelity environment. Success depends on high-quality friction analysis and paired experiments varying only content.
The repo is available on GitHub, and users can start with an npm install. Documentation and a quickstart guide are provided to guide initial setup. The post invites Hacker News users to share their experiences with similar in-house builds or suggestions for improvement.
Source: Hacker News · Summarized by HeadlinesBriefing