Gas Town quietly burns your LLM credits to fix its own bugs

Gas Town, the AI agent tool built by Steve Yegge, is using people's LLM credits and GitHub accounts to fix its own bugs. A GitHub issue opened by user LightOfSeven reveals that the default installation includes formulas that direct local Gas Town instances to review open issues on the maintainer's repository, burn through users' Claude credits to work on fixes, and submit pull requests using their GitHub credentials. There's no opt-in, no opt-out, and no mention of this behavior in the public documentation.

The offending files are gastown-release.formula.toml and beads-release.formula.toml. According to LightOfSeven's investigation, confirmed by Claude's own analysis of the codebase, Gas Town agents pick up "convoys" tracking issues from the maintainer's GitHub issue tracker and spin up "polecats" to work on them. The mayor.md template appears to hardcode GitHub URLs that cause the agents to proactively raise and resolve bugs in the Gas Town software itself. Your credits. Your GitHub account. Spent without asking. The broader conversation often touches on the risks of handing over access, and many experts advise that you should not hand AI agents your API keys.

Hacker News commenters weren't shocked. One noted that the entire Gas Town/Beads stack seems built around a philosophy of "burning tokens" around the clock. Another compared it to the NFT craze, calling Gas Town a "vibe coded mess" that represents the opposite of what production LLM work requires. Yegge's original announcement apparently contained stark warnings, and this behavior seems consistent with that ethos. This effectively quietly burns your Claude credits to fix its own bugs.

This is parasitic computing for the LLM age. The concept dates back to a 2001 Nature paper by Albert-László Barabási, who showed how TCP/IP error-correction could be used to perform distributed computations without explicit permission. Cryptojacking did something similar in 2017-2018 with browser-based Monero mining. SETI@home and Folding@home solved this ethically by requiring explicit opt-in. Gas Town didn't bother.