Top
Best
New

Posted by kristianpaul 10 hours ago

Pi – A minimal terminal coding harness(pi.dev)
289 points | 118 commentspage 3
isagawa-co 7 hours ago|
Interesting approach to planning via extensions. I took a similar direction with enforcement. A governance loop that hooks into the agent's tool calls and blocks execution until protocol is followed. Every 10 actions (configurable), the agent re-centers. No permission popups, but the agent literally can't skip steps.

Open source: https://github.com/isagawa-co/isagawa-kernel

ramoz 9 hours ago||
The way you’re able to extend the harness through extension/hook architecture is really cool.

Eg some form of comprehensive planning/spec workflow is best modeled as an extension vs natively built in. And the extension still ends up feeling “native” in use

indigodaddy 7 hours ago||
too bad I cannot star this.. https://github.com/badlogic/pi-mono/tree/main/packages/codin...
suralind 9 hours ago||
I’ve been testing it for a few days on pretty much clean install (no customizations/extensions) and it’s ok. Not sure if I like it yet.
gtirloni 6 hours ago||
What's a coding harness? Claude Code is a "harness" and not a TUI?
jasonjmcghee 2 hours ago||
The fact that it's a tui isn't particularly relevant. It could be a gui or cli and provide very similar value.

Nearly all of its value is facilitating your interaction with the LLM, the tools it can use, and how it uses them.

ErikBjare 6 hours ago||
If you run Claude Code with `-p --output-format json` it's no longer a TUI, but it's still a harness.
rglover 6 hours ago||
Excited to give this a try, looks really well done.
fred_tandemai 9 hours ago||
Anyone managed to run pi in a completely sandboxed environment? It can only access the cwd and subdirectories
rcarmo 9 hours ago||
I run mine inside https://github.com/rcarmo/agentbox (with https://github.com/rcarmo/webterm)
ac29 7 hours ago|||
Yeah I wrote a small landlock wrapper using go-landlock to sandbox pi that works well (not public, similar projects are landrun and nono).

Note that if you sandbox to literally just the working directly, pi itself wont run since pretty much every linux application needs to be able to read from /usr and /etc

fjk 9 hours ago|||
I’ve been tinkering with Gondolin, a micro-vm agent sandbox.

Here’s an example config: https://github.com/earendil-works/gondolin/blob/main/host/ex...

monkey26 9 hours ago||
I do this with an extension. I run all bash tools with bwrap and ACLs for the write and edit tools. Serves my purposes. Opens up access to other required directories, at least for git and rust.

I think I published it. Check the pi package page.

lukasb 9 hours ago||
But I can't use my Codex plan with it, right? I have to use an API key?
mccoyb 9 hours ago||
You can use your Codex plan with it. OpenAI endorsed it several weeks ago, as far as I remember. That could change, however, but now seems safe.
ac29 7 hours ago||
You can use your Claude or Gemini plan with it too for now, though Anthropic and Google have made it clear this is a ToS violation.
theshrike79 8 hours ago||
Pi makes GPT-5.3-Codex act about on par with Claude easily.

There's something in the default Codex harness that makes it fight with both arms behind its back, maybe the sandboxing is overly paranoid or something.

With Pi I can one-shot many features faster and more accurately than with Codex-cli.

TacticalCoder 4 hours ago|
Naming skills though...
More comments...