@Ryan_Hecht @kzu @GitHubCopilot @pierceboggan @burkeholland Would be nice. Although something generic like .github/plans would work for me too.
English
Sebastian Wenninger
16 posts
@_swngr_ @kzu @GitHubCopilot @pierceboggan @burkeholland Everyone seems to have different opinions on this one! A config option is probably the right move. Would you like to be able to explicitly set the directory in the repo to use?
English
Hey @GitHubCopilot CLI, it's 2026, make that file path a hyperlink please 🙏 /cc @pierceboggan @burkeholland
English
@badlogicgames Love it! I really think adding an agent in the editor itself would add value for teaching/training purposes. Pair-speccing 😬
English
welp, vibed it in an hour. i really only needed single writer, multiple readers/commenters. easy enough.
github.com/badlogic/jot
Mario Zechner@badlogicgames
is there something like google docs, but for markdown? i need a cloud based collaborative markdown editor please.
English
@badlogicgames Not an expert, but maybe WASM could also help here? Sandboxing built in and modular components kinda are its thing as far as I understood.
English
literally been thinking about how i can make a GUI version of pi. easy if you control everything, but extensions make this tricky.
thoughts so far:
- limit to web stack, native apps are dead
- web components with and without shadow root?
- how to transmit over wire in remote case
can't come up with something as easy for extension writers as the gui stuff. kinda sucks :/
English
oh no, thwy made him abandon the cli. what are they doin to our boi.
Peter Steinberger 🦞@steipete
Oh right i realize hell freezes over, we reached the point where app > cli That combined with the speed increase also means less windows necessary, codex goes brrrr now! developers.openai.com/codex/app/
English
@opencode is black still a thing? It's been quiet for a while now
English
@pierceboggan Unify tool names so that custom agents/skills etc. can be used seamlessly.
English
VS Code 🤝 GitHub Copilot CLI
For folks using both together, what should we prioritize improving in the experience?
English
English
@_Evan_Boyle @Grynn Documentation! Feels like X has more info on features than the official docs
English
Me too, switched from claude code to codex (good) & github copilot cli (cheap)
Claude code is bloated af.
kache@yacineMTB
i've officially switched from claude code to codex
English
@_Evan_Boyle Is there a way in copliot cli to pin the standard subagents to specific models?
e.g. if I always wanna use codex for reviews
English
English
@atthatmatt Exactly, "AgentFS" is another way we've talked about this inside the team.
English
We've been working on something internally called "infinite sessions".
When you're in a long session, repeated compactions result in non-sense. People work around this in lots of ways.
Usually temporary markdown files in the repo that the LLM can update - the downside being that in team settings you have to juggle these artifacts as they can't be included in PR.
Infinite sessions solves all of this. One context window that you never have to worry about clearing, and an agent that can track the endless thread of decisions.
Scott Hanselman 🌮@shanselman
Something is cooking in GitHub #copilot
English
@acoyfellow @GeoffreyHuntley So if Ralph is the loop...what's recursion and higher order functions? 🤔
English
Ralph is a primitive, it’s letting people rethink things @GeoffreyHuntley
Arvid Kahl@arvidkahl
Been thinking about the Ralph Wiggum loop a bit. We now have an opportunity to do for Open Source what Folding@home did for protein folding research. Spend a few thousand tokens a day on a single run, trying to improve the performance of a random OSS library/project. Thoughts?
English
@gmickel @OpenAI @RepoPrompt @pvncher @opencode Thanks for the insight! Was wondering whether it is better to have the review in the inner loop so it has context (but also fills more context), or in the outer loop. Maybe even a combination could work...
English
Go for it! I've been able to port it to @opencode with 100% parity, will release that soon.
WRT to the reviews cycle:
Nothing is discardeed! If the reviewing LLM returns NEEDS_WORK or MAJOR_RETHINK, Claude implements the reviewer's feedback and re-submits for review. This happens in the same chat session (preserves context).
This fix→review cycle continues until SHIP or MAX_ATTEMPTS_PER_TASK (default 5). If max attempts hit, task gets marked blocked for human review.
This was part of flow-next before Ralph mode and works really really well.
English
I just published a new version of flow-next that lets you use @OpenAI Codex as the reviewer instead of @RepoPrompt. I still heavily recommend RepoPrompt, @pvncher did great work on the context builder, but this lets all non-mac users use Flow-Next and its Ralph mode!
Gordon Mickel@gmickel
Looked at Anthropic's ralph-wiggum plugin again. Single session, accumulating context, no re-anchoring. This really defeats the purpose. The original vision by @GeoffreyHuntley: - Fresh context per iteration - File I/O as state (not transcript) - Dumb bash loop, deterministic setup Anthropic's version? Stop hook that blocks exit and re-feeds the prompt in the SAME session. Transcript grows. Context fills up. Failed attempts pollute future iterations. The irony: Anthropic's own long-context guidance says "agents must re-anchor from sources of truth to prevent drift." Their plugin doesn't re-anchor. At all. --- flow-next follows the original vision: ✅ Fresh context per iteration (external bash loop) ✅ File I/O as state (.flow/ directory) ✅ Deterministic setup (same files loaded every iteration) Plus Anthropic's own guidance: ✅ Re-anchor EVERY task - re-read epic spec, task spec, git state ✅ Re-anchor after context compaction too (compaction shouldn't happen but if it does, we're set) Plus what we added: ✅ Multi-layered quality gates: tests, lints, acceptance criteria, AND cross-model review via @RepoPrompt ✅ Reviews block until SHIP - not "flag and continue" ✅ Explicit plan → work phases - plan reviewed before code starts ✅ Auto-blocks stuck tasks after N failures ✅ Structured task management - dependencies, status, evidence Two models > one. Process failures, not model failures. Agents that actually finish what they start. github.com/gmickel/gmicke…
English
idk how this fits into git but opencode under the hood has information on which hunks in a diff were ai generated
i feel like this is important info that needs to be surfaced but i also don't know how best to layer this into git
i guess they'd have to be committed separately with a different committer?
English