Ethereal

autonomy-loop 0.8.0: a coding loop that refuses to trust itself. every agent says “the tests pass.” this one proves it by breaking the test. and when it can’t prove its own setup is safe, it refuses to run. how it works: → builder + reviewer pass a git baton. no chat, just a committed LOOP-STATE.md → the bite: reviewer reverts only the fix and demands the test go red. stayed green? the test was lying → brand-new code? it plants a mutant and makes the test kill it → it grades the test, not just the code. assertion-free “green” gets caught → touches money or anything irreversible → it parks for a human → it can’t loop forever: repeats escalate, re-scope → stronger model → park new in 0.8 + planner layer: → four rings of defense, and it earns a trust tier T0 → T3 instead of assuming it → below “hardened” it refuses unattended auto-promotion. provision the rails or it stops → the reviewer must be a separate, live principal. no forged approvals → researchers read the web; injected commands are stripped, only sourced facts pass as data → short on terminals? single-cli mode runs the roles in one, labeled REDUCED-INDEPENDENCE, never hidden → 3-terminal mode (recommended): dedicated Planner researches then grills specs with falsifiable acceptance criteria and goal-ready prompts before the builder touches anything → 4-terminal power mode adds a Researcher feeding an idea pool install (it’s a claude code plugin): → claude plugin marketplace add github.com/inferencegod/a… → claude plugin install autonomy-loop → /autonomy-init, then run /autonomy-loop:builder and /autonomy-loop:reviewer (add planner for full setup) in terminals on /loop 600 two (or three) terminals. a builder, and a reviewer that trusts nothing. one repo. nobody driving.

@5fakb48 @ClaudeDevs Burn them tokens x.com/inferencegod/s…

This is my first month paying and the resets are not the same as Codex lol, as people are saying the trend is clearly you guys resetting the weekly super close to the users actual weekly renew ( my case too lol ). For the love of all that's good in this world, either add the storage of a reset like Codex, or use the reset mid weed... Sigh

Update: we've gone ahead and reset 5-hour and weekly usage limits for everyone, across all plans. Enjoy your weekend!

@ClaudeDevs best way to burn them tokens til sunday 🫪🫪x.com/inferencegod/s…

the same-folder clash is pain, that’s exactly why mine run on separate git worktrees coordinating through a committed baton instead of a shared working tree. but that solves build-time collision, not your problem. “blocked for an hour” merging into main is almost always branch protection, not the agents. the usual hour-eater is “require branches up to date before merge”: every merge puts the others behind main, so they each rebase and re-run checks, one at a time. turn that off or put a merge queue in front so it handles the updating. if it’s not that, it’s secondary rate limits from a shared token in a tight merge loop, fixable with per-agent short-lived tokens. no agents.md fixes either one, it’s repo settings.

Didn't work. $500 to anyone who sends me an agents.md file that solves my issue. My agents should be able to deploy multiple releases (merge into main) in quick succession without getting completely blocked for an hour.

Codex feels good again 🥹

@MatthewBerman submitted /autonomy-loop ! x.com/inferencegod/s…

Launched the Loop Library Skill Now your agent can find existing loops or help you create loops from scratch. github.com/Forward-Future…

@martintechlabs single cli mode added 🤞

@inferencegod Why not just use subagents?

i don't feed my agent tasks anymore. when the backlog runs dry, it researches and invents the next feature itself, then builds it. and it polices its own work before i ever see it. autonomy-loop v0.5.1: → self-feeding: empty backlog? it proposes the next feature and keeps going, no prompt from me → the bite: it reverts its own fix and reruns the test. stays green? it caught nothing, rejected → self-mutation: it mutates its own changed lines so weak tests get caught before handoff → circuit breaker: it parks to me instead of looping forever → branch protection: it can never touch prod or edit away its own gates → upgrading is one command: /autonomy-upgrade → red-teamed, 77 tests green two terminals. a builder, and a reviewer that trusts nothing. one repo. nobody driving. free, mit, 151 people already running it. /plugin marketplace add github.com/inferencegod/a… /plugin install autonomy-loop@autonomy-loop

@0xDesigner Idk but convert this into glm and run it for infinity you are welcome brother x.com/inferencegod/s…

to the people running GLM 5.2 locally, how are you doing it? i’m starting to seriously consider running a local model on my mac mini. what do i need to know?

@omarvvvr check this out bro , i honestly cannot relate x.com/inferencegod/s…

genuine question: AI was supposed to make us work less so why does it feel like everyone is working 10x more now?

biggest release yet for autonomy loop incoming

@rmsm1th 😳😳 challenge accepted

@inferencegod You’ll be a billionaire if you figure out how to do this for non coders

@theo This catches a lot before a strong point of failure x.com/inferencegod/s…

One of these loops went so bad that I made Codex write an apology document

there’s actually a third move: use both, but let the deterministic one bind. in my loop the LLM critique advises, but a deterministic gate has the final say. the reviewer reverts the builder’s fix and confirms the test actually goes red. if it doesn’t, the test was lying, and no amount of “looks good to me” lets it pass. LLMs are great judges right up until they’re judging their own work. that’s where the deterministic floor matters. autonomy-loop, open source: github.com/inferencegod/a…

There are two kinds of loops: > deterministic (ex: when site speed less than 50ms) > LLM as judge (ex: when the website is fast enough) LLMs are becoming fantastic judges. This opens up significantly more knowledge work to be looped.

@theo @kdy1dev just me and michael fr x.com/inferencegod/s…

@kdy1dev You need more subagents clearly

/autonomy-loop

@jonas_nelle gonna bring this over to cursor soon :-) COMPUTA AUTOMATE THE AUTONOMY x.com/inferencegod/s…

Automating your repetitive work has never been easier - just run /automate in Cursor

@NebulaAI @ElevenLabs @grok Niceee!

Create any video you can imagine. - Ask your agent to make a script - Generate image stills - @ElevenLabs for VO and SFX - @grok, Veo 3.1, Seedance 2, Kling 3 and more - ffmpeg pieces it all together All within our sandbox. DM us for the prompt we used. Full ad at end👀

yeah you read it right, they coordinate through committed git state, not direct chat. the back-and-forth-forever thing is real for pure two-party loops, but two things stop it here. the reviewer isn’t just voting, it runs a mechanized gate (revert the fix and confirm the test goes red, coverage ratchet, patch coverage) so the verdict is pass/fail against real code, not an opinion that can ping-pong. and there’s a no-progress breaker: consecutive waves with no tree change park the loop to me instead of looping forever. your “going too deep before feedback” point is fair though. i’ve actually watched the reviewer pull the builder back from a bad path a few times, so commit-then-review catches more than you’d think. real-time sharing getting feedback earlier is a real edge, i won’t pretend otherwise. 0.6.0 (live now) already adds a planner that grills the spec before the builder builds, basically the third perspective you’re describing, just upstream instead of a runtime vote. the coordination stuff you’re hitting is exactly what i’m building next, more coming tonight !

Very cool, will check it out. Looks like you have 1 builder and 1 reviewer and they don’t directly communicate is that right? Does that ever get stuck for you? Or maybe I’m misunderstanding but I’m my experiments I’ve seen the reviewer and builder go back and forth in perpetuity without a third vote. Found real time context sharing to be useful in that regard as well as opposed to one builder going down a path too deeply before getting feedback.

Just launched Loop Library - a curated list of agent loops you can use right now. Find loops, submit your own, tokenmaxx!! signals.forwardfuture.ai/loop-library/

@DavidKPiano inputs fsm loop x.com/inferencegod/s…

Everyone hyping loops right now is going to absolutely lose their minds once they learn about state machines

@willwashburn @MatthewBerman This has the gate for that level of work very interesting idea x.com/inferencegod/s…

@MatthewBerman imo loops still leave the human at the center. instead give an agent a set of high level intents and constraints, and have it write the loops and exit conditions. loop the loop, if you will. without the right guardrails ends in slop for sure, but with them...