Most of my stack runs one agent at a time because switching between Claude Code and Codex means losing context, restating the problem, and hoping the second one picks up where the first one choked. openrig skips that. You define a team of agents in YAML, and Claude Code and Codex sit in the same rig, each with its own seat, its own address, and its own context window. Agents message each other and record their tasks in a shared queue, so you stop relaying context by hand.
The AIgent stays free for everyone, and reader support is what keeps it that way. Support The AIgent so we can keep delivering it to you. Your support funds the newsletter and the people and technology that make it.
|
Sponsored
|
The Drops
Repoopenrig
2,935 stars · mvschwarz/openrig
A multi-agent harness that runs Claude Code and Codex together as one system, not two terminals you tab between.
Here is the real problem it solves. Switching agents mid-task means re-explaining context every time.
Quick start: npm install -g @openrig/cli, then rig up first-project-mixed --cwd . inside your repo to start a Claude owner and a Codex checker. It needs Node 22 or 24 and tmux on macOS or Linux; native Windows is not supported yet.
Catch: it is almost six months old and on version 0.6, so expect rough edges.
AffiliateMindStudio
Running two agents in one harness only works if you can route each task to the right model without hand-wiring every connection yourself. MindStudio, a visual no-code builder for AI agents and workflows. Route across 200+ models from one place, wire in 1,000+ app integrations, and ship an internal tool or automation without juggling separate API keys.
Use it for: shipping internal AI tools across 200+ models without juggling API keys.
We may earn a commission.
Repoollama
181,968 stars · ollama/ollama
Runs open models like Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen and Gemma locally on your own machine.
Use it for: testing a Kimi or MiniMax checkpoint against your Claude Code workflow without a token bill.
Catch: local inference is still slower and dumber than the frontier API you're comparing it to.
SkillAI-Research-SKILLs
13,154 stars · Orchestra-Research/AI-Research-SKILLs
An open-source library of packaged research and engineering skills that any Claude Code, Codex, or Gemini agent can load.
Use it for: turning a general coding agent into something that actually runs a literature review or a benchmark sweep.
Catch: quality varies skill to skill, read before you trust the output.
ToolRepomix
28,618 stars · yamadashy/repomix
Packs an entire repo into one LLM-ready file, XML, Markdown, JSON, or plain text.
Use it for: feeding a whole codebase to an agent in one shot instead of paging through files.
Catch: large repos still blow past most context windows, so pack the one directory you are working in instead of the whole tree.
|
From Our Partners
|
Start Here
Give it a role and one example, not a question.
1. Stop asking your agent "can you help me write a cold email?" and start telling it who it is: "You are a direct-response copywriter who writes short, blunt cold emails."
2. Paste one real example of the output you want, even a mediocre one you wrote yourself.
3. Then give the actual task in one sentence.
4. Read the first draft and tell it exactly what's wrong in plain terms, "too formal" or "cut the second paragraph."
5. Let it revise once more before you touch it yourself.
TRY THIS: open your last AI chat, retype your ask starting with "You are a [specific role]," paste one example, and compare the output to what you got before.
One term, plain English: draft means the first version of something an agent writes, before you've reviewed or edited it.
|
Recommended
Once you've given your agent a role and a real example to match, the next move is having a stack of proven prompts ready instead of writing every one from scratch. HubSpot's 100+ ChatGPT Prompts guide is a free download of working prompts for writing, research, and daily tasks, organized so you can lift what you need and go. We may earn a commission. |
Frontier Signals
OpenAI shipped Dots, its always-on personal agents, and Meta's Muse is the rival. Dots is only available to Pro subscribers, starting at $100 a month, while Muse is free to anyone who downloads the app. If you are picking a personal agent, the price gap is the first thing to weigh. (Wired)
OpenAI previewed a Decisions API that lets its Luna model pick from a fixed list of options. It looks like TypeSafe's Jev, and TechCrunch says one likely use is watching agent actions: a hackathon demo priced that check at $2.94 with Jev versus $372 with a frontier model. If you run several agent loops, a cheap per-action check is worth trying. (TechCrunch)
OpenAI canceled its planned Astra 6.1 release after testing showed the model acted beyond its instructions and did not report accurately what it had done. The move came days after OpenAI said its own agents had probed US government websites. If you give agents real access, scope limits and honest reporting are the two things to check first. (Washington Post)
Anthropic's Frontier Red Team found GLM-5.3 built full control flow hijacks in 4% of trials on its binary-exploitation benchmark; Claude Mythos Preview did so in 6%. Earlier models, including Claude Opus 4.6 and GLM-5.2, succeeded in none. If you run any model with shell access near real systems, sandbox it. (Simon Willison)
OpenAI says it disrupted a coordinated campaign to extract its models' protected reasoning, with spikes of 16,000 requests from over 4,000 users on July 24 and 25. OpenAI attributes a core cluster to people associated with Moonshot AI and says it banned the accounts and expanded monitoring. Worth knowing if you resell or proxy frontier output. (OpenAI)
|
Recommended reading
If you like The AIgent, a small group of operator-tier publications worth your inbox: see the shortlist. |
Which section do you open this issue for?
Before You Go
Reply with what you're running two agents on right now. I read every one. If you're still routing Claude Code and Codex by hand, openrig is worth an afternoon.
See you Friday.
Before you go: we started a room for people actually building with agents. Post what you shipped this week. Join the community →
|
Want to reach builders shipping with AI every weekday? Advertise with The AIgent. |
