Gas Town
Gas Town Hall (Steve Yegge) · Orchestrators & meta-harnesses
The Register / Orchestrators & meta-harnesses / Gas Town
Register entry · Orchestrators & meta-harnesses · updated 26 Sept 2026
Gas Town (gt) coordinates many coding-agent CLIs - Claude Code, Codex, Copilot, Gemini, Amp and others - across projects ('rigs'). A Mayor agent plans work, ephemeral worker agents ('polecats') run in git worktrees, watchdog roles restart stuck agents, and a per-rig merge queue batches and verifies merges. Work state persists in Beads, a git/Dolt-backed issue tracker. Development has largely moved to Gas City, an orchestration SDK extracted from Gas Town.
Deprecated. In maintenance mode: a project collaborator wrote on 2026-05-01 that Gas Town is 'staying focused on infrastructure and reliability fixes only' and pointed users to its successor, Gas City. Last release v1.2.1 on 2026-06-06; last commit on main 2026-07-23. Repo moved from steveyegge/gastown to gastownhall/gastown (old URL redirects).
Gas Town Hall (Steve Yegge) · Orchestrators & meta-harnesses
| Price | Free |
|---|---|
| Use it via | CLI |
| Runs on | macos, linux |
| Licence | MIT |
| Updated | Evidence collected 26 Sept 2026 |
Weights for Orchestrators & meta-harnesses: adoption 40%, experts 30%, crowd 30%. Pillars without enough evidence are left out and the others re-weighted. The method.
| Adoption & momentumweight 40% | 48 | GitHub stars 18k → 69 · 0 releases in 90 days → 0reach × 0.7 + momentum × 0.3 |
|---|---|---|
| Expertsindependent reviews · weight 30% | 50 | 4 independent reviews (1 positive · 2 mixed · 1 negative), averaged toward 50 for small samples. |
| Crowdcommunity sample · weight 30% | 44 | 11 dated posts from reddit, hackernews: 3 positive · 3 mixed · 5 negative (positive + ½ mixed) ÷ n, averaged toward 50 for small samples |
| Rating | 47 | medium confidence (3 of 3 pillars, 11 community posts). |
Evidence checks. 1 GitHub issue not counted as sentiment. How evidence is checked.
Seven dimensions graded A–D from documented facts. Overall: B.
Built-in agent presets launch with approvals disabled: Claude Code with --dangerously-skip-permissions, Codex with --dangerously-bypass-approvals-and-sandbox, Amp with --dangerously-allow-all. Users can define custom agent commands, so safer modes are opt-in. source
Agents work in per-rig git worktrees but run with the user's full account access; no documented allowlist or scoping model beyond worktree layout. source
Runs locally (or in the user's Docker container); work state is stored in local git/Dolt databases. No vendor cloud. source
Local tool with no vendor data collection documented; no retention or training policy exists to grade. source
No CVEs or published advisories. April 2026 issue #3649 (253 points on HN) reported that bundled release formulas could have users' agents work on Gas Town's own issues using their LLM credits and GitHub accounts; it was closed on 2026-05-01 by a collaborator citing maintenance mode, with no documented code change, and the formula files remain in the repo. Whether it triggers without user direction was not verified, so no grade. Several internal security fixes (input sanitisation, gt-sec-001/002) were made in March 2026. source
No attestations; community open-source project. source
MIT licence with a dated Keep-a-Changelog CHANGELOG.md. source
4 named independent reviews found; dated reviews from the last 12 months are scored.
It feels like all of the friction has been taken out of managing multiple agents.Mike Lady (Enterprise Vibe Code) · 3 Jan 2026 · newsletter.enterprisevibecode.com
Gas Town is a glimpse into the future: a dark, grim future where we are still catching up on the tooling.John McBride · 16 Jan 2026 · johncodes.com
The biggest flaw in Yegge's creation is that it is poorly designedMaggie Appleton · 23 Jan 2026 · maggieappleton.com
When the cheap gas goes away, everyone will leave Gas Town.slowlywilliam · 12 Apr 2026 · slowlywilliam.com
11 coded posts from reddit, hackernews.
the core ideas are right: persistent tasks, isolated workers, smart hub. the execution is early.r/ClaudeCode — Spent 2 weeks running multiple claude code agents in parallel with gastown · 3 Feb 2026 · reddit.com
TBH this post still reads like a clown show.Gas Town: From Clown Show to v1.0 · 14 Apr 2026 · news.ycombinator.com
Gas Town really feels not just vibe coded but also vibe designed.Gas Town: From Clown Show to v1.0 · 14 Apr 2026 · news.ycombinator.com
I haven't used gas town in a month or so, but a stable gas town sounds very valuable.Gas Town: From Clown Show to v1.0 · 14 Apr 2026 · news.ycombinator.com
Gas Town is the exact opposite of what is needed to actually do useful things in prod.Does Gas Town 'steal' usage from users' LLM credits? · 15 Apr 2026 · news.ycombinator.com
Orchestration finally clicked with me when I tried out Gas Town six months agocomment · 22 Aug 2026 · news.ycombinator.com
Something can be ridiculous and also conceptually interesting / intelligent.comment · 22 Aug 2026 · news.ycombinator.com
ok, but when steve made gas town, it didn't actually do anything but build more gas town.comment · 25 Aug 2026 · news.ycombinator.com
As published, checked 26 Sept 2026.
| Plan | Price | Notes |
|---|---|---|
| Open source | Free | MIT-licensed CLI; users pay for the underlying agent subscriptions/API usage. |
Last commit on main (merge-queue and post-merge verification fixes). source
v1.2.1: shell integration no longer prompts in arbitrary repos; routing and mail fixes. source
v1.2.0: daemon distinguishes Claude usage limits from crashes to avoid restart loops. source
Collaborator states Gas Town is in maintenance mode; broader work moves to Gas City. source
v1.0.0 released. source
Other ranked entries in Orchestrators & meta-harnesses.
Every figure above traces to one of these, observed up to 26 Sept 2026.