t3-pstack: Porting the pstack Agent Framework to T3 Code

X postChristian Bager Bach HoumannTool

Christian Bager Bach Houmann has ported Lauren Tan's pstack framework to T3 Code. The tool directly integrates pstack with T3 Code's native MCP tooling, enabling multi-model reviews across Claude, GPT, and Grok.

Key points

  1. Based on Michael Denyer's pstack-claude fork, dispatching subagents, fan-outs, schedules, and panels via T3 Code MCP tools.
  2. Allows running Claude, GPT, and Grok reviewers side by side in panels (such as interrogate, arena, and architect) from Claude Code or Codex threads.
  3. Designed specifically for T3 Code, with the author recommending T3 Code Nightly for the smoothest experience.
  4. Includes features like the 'ketchup' skill, which summarises intermediate action items from long-running agents without requiring full trace reviews.
  5. Maintained as a dedicated plugin namespace (t3-pstack) that avoids collisions with upstream pstack-claude installations.

Why it matters

Developers leveraging multi-agent workflows in T3 Code can use standardised pstack processes. This facilitates automated cross-model code reviews and orchestration directly within their development environment.

Analysis

The project illustrates the growing shift toward harness-specific multi-model orchestration. Rather than maintaining a generic cross-tool layer, t3-pstack tailors agent routing and panels directly to T3's native MCP capabilities. Houmann emphasizes an asynchronous workflow where developers set goals upfront, let long-running agents execute independently, and only intervene during planning and final pre-shipping checks.

Original post

Christian Bager Bach Houmann

@chrisbbh · 13 October 2026

I ported @poteto's pstack to @t3dotcodes. Feels absolutely incredible to use. It's native T3, meaning we use the T3 Code built-in tools and tune the stack entirely to t3. I'm trying to keep it faithful to pstack while optimizing it for t3. t.co/x5SIy5Agi8 t.co/VbsJF4iSBx

760 likes38 replies1.1k bookmarks42.1k views

View on X
More posts in the thread (11)
  1. @ZaharovAndrew @poteto @t3dotcodes Yes, that gives the best experience. I'm building this on nightly. Generally nightly is pretty stable for 'nightly' - haven't had any major issues
  2. @hushyfu @poteto @t3dotcodes Because I used it myself in Claude. It was OK but I’ve made substantial improvements since and focus on being faithful to the upstream - but still, things need to be custom for T3. Would love to hear any feedback and will quickly ship improvements
  3. @zhxnrss @poteto @t3dotcodes I haven’t used nor taken more than a cursory look at open-pstack, but this is specifically tailored to and optimized for t3. No compromises. I’m not balancing tradeoffs or trying to fit skills to multiple between multiple harnesses. pstack is tailored for Cursor. This is for
  4. @drag00n3333 @poteto @t3dotcodes It's mostly me writing a great prompt that explains what I want and a direction. The idea is to get out of the loop outside critical moments, which is usually when defining what needs to get done, and before shipping it to users. I.e. you need to set the direction, and you need
  5. @drag00n3333 @poteto @t3dotcodes But I do like using the new "ketchup" skill. Long running agents often write action items or steps in interim messages, and I don't always catch those. So it's nicer to have it catch me up rather than having to read the full trace each time.
  6. @YounesAka @poteto @t3dotcodes This repo is specifically tuned for t3 and I'm working heavily on it, so I'd prefer to be in control of what comes out. Running tons of evals and have a huge data flywheel to ship continuous improvements. Also, so far this repo updates faster with upstream. Huge respect for
  7. @Jordy_vD_ @poteto @t3dotcodes that'd be great but alas my excitement to share my tools and expertise remains unpaid
  8. @dakdevs @poteto @t3dotcodes I try to add more accounts if I feel bottlenecked by tokens
  9. @hushyfu @poteto @t3dotcodes Thanks! Brigade and landing are pstack-t3's own additions. In my experience, models can just do this on their own now. Prompt attached has been running for days and it just does all of what you mention on its own. Sorry to redact, but you get the sense that it has a massive t.co/NmBWrcEbG3
  10. @CastanedaEdward @dakdevs @poteto @t3dotcodes It's not really a problem to switch context, but I do try to keep threads to a single account if possible due to thinking traces being stored remotely and switching busting that
  11. @dmooralesg @dakdevs @poteto @t3dotcodes Highly recommend looking into CLIProxyAPI ... and try to use a home IP if you do, no hosting on a VPS unless you have a home tailscale exit node it can proxy through
Selected replies (5)
  • @ZaharovAndrew @chrisbbh @poteto @t3dotcodes Tempted to try it out. Do you need to be in T3 Code Nightly for best results? I think the T3 MCP is still not in the stable channel.
  • @CaptainFirmware @chrisbbh @poteto @t3dotcodes Looked at all 4 Yours is the smoothest for Claude/Codex reviews: t.co/woNuTnCpbj — just one step behind Cursor and needs hand updates. This one matches newest pstack exactly and runs on all three I use — grok, codex, and opencode: t.co/I89sGzPjPs It tracks the
  • @YounesAka @chrisbbh @poteto @t3dotcodes why not a PR on michael's repo?
  • @Jordy_vD_ @chrisbbh @poteto @t3dotcodes Do you get paid every time you mention t3
  • @hushyfu @chrisbbh @poteto @t3dotcodes Really appreciate your work! I currently lean towards creedants/pstack-t3 because of its brigade orchestration, isolated worktrees, cross-model reviews and merge queue. How does your port compare in those areas? My main concerns are upstream parity, sync reliability and context

Mentioned links

Links and tools from this entry

Summary generated by AI (Gemini 3.8 Flash, 14 October 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Repository:chhoumann/t3-pstack

    t3-pstack: Multi-Model Orchestration for Coding Agents

    t3-pstack adapts Lauren Tan's pstack and Michael Denyer's pstack-claude for T3 Code, dispatching subagents, review panels, and fan-outs across Claude, GPT, and Grok via T3's MCP tools.

    172starsJavaScript

    AI & LLMs· Tool

  • Repository:pingdotgg/t3code

    T3 Code: Unified Control Surface for Local AI Coding Agents

    T3 Code is an open-source agent harness control surface that lets developers interact with and control local AI coding agents via desktop, web, and mobile applications.

    21.6kstarsTypeScript

    AI & LLMs· Tool

  • Repository:FeiZhuLulu/Agent-Bridge

    Agent Bridge: Orchestrating Local Coding Agents via MCP

    Agent Bridge is an open-source Python tool connecting local coding agents using the Model Context Protocol (MCP). It allows users to interact with a single coordinator agent that delegates tasks to multiple specialized CLI worker agents.

    31starsPython

    AI & LLMs· Tool

  • Repository:Leonxlnx/tastecode

    TasteCode: Local Desktop Workspace for AI Coding Agents

    TasteCode is an open-source desktop application for Windows and macOS that combines multiple AI coding agents such as Claude Code, Codex, and Grok into a unified interface. The tool follows a local-first security architecture and includes a dedicated Design Mode for editing web interfaces.

    367starsTypeScript

    AI & LLMs· Tool

  • Video

    Video:CURT

    Agentic Development Stack: Harnesses, Models, and Automated Workflows

    Software developer Ross details his updated engineering stack for building agentic AI applications. He explains his use of harnesses like T3 Code, Cursor, and Claude Code, his preferred models Opus 5.5 and GPT-6 Astra, core frameworks including SvelteKit, Convex, and Effect, and an automated Git worktree pipeline featuring Greptail review loops.

    AI & LLMs· Talk

  • Repository:EveryInc/compound-engineering-plugin

    Compound Engineering: 33-Skill Plugin for AI Coding Agents

    Compound Engineering is an open-source plugin for AI coding agents designed to structure development workflows and systematically preserve learned context. It runs on 14 platforms, including Claude Code, Cursor, and Codex.

    24.9kstarsTypeScript

    AI & LLMs· Tool

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.