superpowers VS gastown

Meanwhile, the ecosystem materialized. obra/superpowers shipped a complete software development methodology built on composable skills — 1,276+ stars and growing. The cobusgreyling/loop-engineering repo cataloged patterns from Osmani and Cherny into a practical reference.

Projects like Superpowers proved that agents can follow structured methodologies — brainstorm before coding, write tests before implementation, review against specs before declaring success. The skills are mandatory workflows, not suggestions. The agent checks for relevant skills before any task.

If you're looking for complementary skills and plugins then have a look at Addy Osmani's Agent Skills or Superpowers, both provide essential coding assistant skills like Test-Driven Development (TDD). OpenSpec provides consistency, and your workflow can evolve around it.

Before any implementation, I plan. I use the brainstorming skill from Superpowers to think through the approach, then Grill Me - a separate skill that probes for contradictions, gaps, and missing assumptions, question by question. Once I'm satisfied, I save the result as a PRD and move to writing-plans (also from Superpowers), which produces a detailed implementation plan: class names, properties, architecture, tests.

These don't replace the spec; they govern how the agent acts on it. Superpowers uses guided Q&A to clarify intent, then runs sub-agents behind a verification-before-completion gate. GSD manages context in waves for solo developers. HVE Core runs an RPI loop: Research, Plan, Implement, Review.

> It would be extremely cool to be able to write one or two lines of prompt in my harness, and have a light model iterate with me a few times writing/proposing requirements, guidelines and explanations, refining the prompt until it's ready to be sent to the actual LLM.

I feel like the vast majority of AI-using coders already do this via skills suites like Superpowers (see /superpowers:brainstorming), no? https://github.com/obra/superpowers

I sometimes use Superpowers skills such as writing-plan and writing-spec. The Superpowers brainstorming skill stores design specs under docs/superpowers/specs/YYYY-MM-DD--design.md, and the writing-plans skill stores implementation plans under docs/superpowers/plans/YYYY-MM-DD-.md. (GitHub, GitHub)

obra/superpowers – GitHub

I mean, Google already has Mu Zero, which Im willing to bet has evolved quite a bit in private because if anything is going to get us closer to actual AI its that.

Realistically, one can build a AI capable of reasoning (i.e recurrent loops with branches) using very basic models that fit on a 3090, with multi agent configuration along the lines https://github.com/gastownhall/gastown. Nobody has done it yet because we don't know what the number of agents is required and what the prompts for those look like.

The fundamental philosophical problem is if that configuration is possible to arrive at using training, or do ai agents have to go through equivalent "evolution epocs" to be able to do all that in a simulated environment. Because in the case of those prompts and models, they have to be information agnostic.

On the other extreme we've got things like Gas Town introduced by Steve Yegge where you just let a swarm of agents rip, barely read code at all and focus on making that setup produce the desired outcomes.

From the most recent comment, looks like this is a bug, triggered by the system inadvertently activating an internal release tool [0]. Still a pretty wild bug, but not as dramatic as the title suggests. Which is kind of unfortunate honestly, the chaos of every gas town instance automatically contributing to itself would be beautiful to see.

- https://github.com/gastownhall/gastown/blob/main/internal/fo...

I'm looking forward to trying this. I've had a positive but high-variance experience with Gastown[1], which is in the same genre. I hope that Scion does better.

My main complaints with Gastown are that (1) it's expensive, partly because (2) it refuses to use anything but Claude models, (3) I can't figure out how to back up its beads/dolt bug database, which makes me afraid to touch the installation, and (4) upgrading it often causes yak shaving and lost context. These might all be my own skill issues, but I do RTFM.

But wow, Gastown gets results. There's something magic about the dialogue that happens between the mayor and the polecats that leads to an even better experience than Claude Code alone.

1. https://github.com/gastownhall/gastown/

Hi HN,

With the emergence of the Model Context Protocol (MCP), we are rapidly connecting large language models (LLMs) to critical infrastructure, APIs, and local files. However, the current standard assumes an extremely trusted 1:1 topology. On the other hand, newer agent-based worms (such as the BYOMCP exploits or OpenClaw) demonstrate that malicious payloads can dynamically overwrite an agent’s context window.

I have just submitted an article to arXiv (which is currently in the queue under cs.NI) that uses the Rice theorem, Kolmogorov complexity, and recent cryptographic proofs to demonstrate why attempting to solve this problem using asymmetric “security wrappers” is unfeasible from a mathematical standpoint. Guaranteeing 100% AI safety is computationally undecidable.

Instead of static filters, the article proposes a reactive topological defense mechanism: the Byzantine MCP Router (BMR). It acts as middleware that establishes a 1:R:N topology. Rather than relying on a single model, it sends the MCP tool request via multicast to several different base models.

To ground the theory in real-world engineering, the paper includes a case study on Steve Yegge's recently released "Gas Town" orchestration framework for Claude Code (https://github.com/steveyegge/gastown). It demonstrates how persistent Git-hook memory in standard hierarchical topologies creates a permanent incubator for agentic worms if a worker agent's context is hijacked.

Key concepts:

If you want the full picture of how Gas Town works, Steve's original blog post is worth every one of its 25 pages. And the repo is open on GitHub if you want to dig into the code.

The other main thing I think it's worth mentioning is Gastown (another of Steve Yegge's projects). I have been exploring this a bit, and the concept has a ton of merit. I don't know what the form factor will be but I hope that the next time I write a post like this it will be about how I went from working with 5-8 agents effectively, to managing swarms of agents that we don't even bother counting. But that's a topic for a while out, maybe summer 2026.