Gas Town, Steve Yegge's autonomous AI agent system, is accused of silently spending users' LLM credits and GitHub credentials to fix its own bugs. Issue #3649, opened by user LightOfSeven, alleges that two default configuration files (gastown-release.formula.toml and beads-release.formula.toml) cause local installations to scrape open issues from the maintainer's repository, burn through users' Claude Code credits working on those issues, and submit pull requests back to the upstream repo using users' own GitHub credentials.
The issue report is blunt. According to the filing, there's no mention of this behavior in the public README or documentation. No opt-in and no warning. Your AI agent could be spending your money and your GitHub identity on someone else's open source project, and you'd never know until you checked your usage logs. The report quotes Claude's own investigation into the behavior, confirming that polecats (Gas Town's worker agents) were actively working on issues from the maintainer's tracker and submitting PRs upstream.
Steve Yegge calls his design philosophy 'Vibe Coding.' Software is a living autonomous entity. The Beads framework is built to be token-intensive, prioritizing rich context and rapid iteration over efficiency. So every Gas Town installation acting as a contributor makes sense, from that autonomous agent angle. Hacker News was brutal. One commenter compared it to the NFT craze, calling it a 'vibe coded mess' that represents the opposite of careful LLM usage.
The issue requests that this upstream contribution behavior be removed from default installs and made opt-in only. It currently carries a 'needs triage' label with no maintainer response.