📣 What if every open issue had a Codex agent? That’s the idea behind Symphony, an open-source agent orchestrator for Codex that turns task trackers into always-on systems for agentic work, letting humans focus on review and direction.

Day 2 requesting Codex app for linux

Nice! I’m a big fan of Symphony. I’ve been using it with a top layer orchestrator (Codex or Claude Code I prompt with), which initiates Symphony and @linear, and which also handles (itself or with its own agents) integrations and any other issues. I made this skill with Codex for the orchestration layer, which also sets up a self-learning file system for helping improve Symphony runs over time. Any tips to further make use of Symphony?

Thanks to Symphony, I’ve created and opensourced my Lacrimosa. I’m adapting it Codex agents now.

Have used Symphony a bunch and it's awesome. Only downside is that it's very token hungry.

Game changer 🔥 Finally, agents that actually ship PRs while we sleep. Symphony + Codex on Linear sounds like the end of ticket backlog hell. Anyone tried self-hosting it yet?

Very normal evolution for the humble to-do list. First it nagged you. Now it may acquire a small workforce and ask for managerial oversight

We run 12 agents on scheduled heartbeats already. The hard part isn't spawning an agent per issue. It's preventing agent 7 from undoing what agent 3 just shipped. Orchestration without conflict resolution is just parallel chaos.

Very interesting. I recently crated a hack to use Linear as the backbone for letting Codex and Opis handle open tickets and communicate with each other in comments and QA each others work, so symphony definitely sounds interesting.

Why not build Symphony into Codex app itself?

Honestly, I cannot state how many open-source repos have something like this. Having this for Codex is always pretty cool. Haven't tried it yet, but I think it's worth a whirl. Perhaps even if we can have AI generate task for it! (Feed AI into the AI)

We released this last week 😛try Broccoli which works on Linear

the harness was last quarter's story. orchestration is this quarter's. symphony, agent studio, claude code's task management, anthropic skills, langgraph. five orchestration layers shipped in 90 days. the product isn't the agent anymore. it's the thing that decides which agent runs which issue and when.

"... an open-source agent orchestrator for Codex that turns task trackers into always-on systems for agentic work, letting humans focus on review and direction." Use common language that everyone can understand.

Is this…new? I remember seeing this repo months ago. Is it polished, easily set up, getting built into Codex? Cool open source project. Better if built into codex as a feature.

@DanielleFong It is. I built my own and man oh man this is seriously the way. Actually it’s also the way to write software that agents use, just build agent capability right into the app, and the agent just uses it from day 1

How do you handle the orchestration when agents hit conflicting priorities across multiple repos? We've seen similar automation break down when the volume gets high and context switching becomes expensive.

@cursor_ai @ericzakariasson can we get something like this for cursor pls

Impossible with your rate limits. A feature only for yourself. Stop dreaming

If you're aim is to put out a product that Anthropic can't afford to put out...

Surprised the cracked engineers at Network School - like @jinglescode - already shipped & open-sourced this months ago… only for OpenAI to drop it today like it’s a brand new frontier breakthrough. Man, is @ns cooking harder than OpenAI? 🔥

With the speed AI executes on your ideas, I don't think this truly becomes powerful before they can add items to the todo list themselves. If you review the output anyways, why not let it explore more possibilities

Looks like Notion for Agents. Pretty cool!

what if limits were as good as before 🥺 I know the model is expensive but…

Interesting, I developed "Accord" to meet my needs. GPT model acts as a reviewer of Claude's work. Build a meta-plan, development cycles and implement them. Every level is going though GPT's review before it is approved. Open Source:

Interesting, looks like the future of work. I actually built a version of this where you can use Codex, Claude Code, Gemini CLI, and OpenCode and you don’t have to manage it:

the question after 'what if every open issue had an agent' is 'what if every closed issue had an agent that closed it for the wrong reason and nobody noticed.' review only catches the obvious failures; the quiet wrong-fixes are how systems rot.

It is cool! Symphony is great for teams. But what about the solo founder with 100 agents? See what we're building:

Yoo Iove this, I hope codex will add also memory to codex like hermes agent has. it would be such a game changer.

Every open issue gets a Codex agent. The backlog is fully staffed. Still a backlog.

now let's add a real time group chat.. and then a openai equivalent version of github to store and train code on.. and then a openai marketing section for people to promote their apps and websites, vetted by codex ... github is smoked

Feels like the big thing is not one smarter agent, but many small agents with a human checking the work. I think a lot of teams will work like this soon.

It would be really cool if you could embed this technology into the Codex app, with it fully managing it. A "project" would go from something that holds my threads, to a literal project management system.

Idea for Codex: In addition to Plan Mode, could you add a simple task Kanban board? Plan Mode could create larger tasks, break them into smaller ones, and keep the original plan context attached to each task. Would make bigger coding sessions easier to manage and resume 😀😅

Been running scheduled Claude Code agents on cron for content + trading pipelines. Bottleneck flipped a while back — not agent throughput, but the human review queue. How does Symphony triage when many agents finish at the same time?