Alternatives
Products that do what Terminal-Bench-RL: Training long-horizon terminal agents with RL does
After training calculator agent via RL, I really wanted to go bigger! So I built RL infrastructure for training long-horizon terminal/coding agents that scales from 2x A100s to 32x H100s (~$1M worth of compute!) Without any training, my 32B agent hit #19 on Terminal-Bench leaderboard, beating Stanford's Terminus-Qwen3-235B-A22! With training... well, too expensive, but I bet the results would be good! *What I did*: - Created a Claude Code-inspired agent (system msg + tools) - Built Docker-isolated GRPO training where each rollout gets its own container - Developed a multi-agent…
- 1IR
Jul 2026 · github.com
- 2OA
Scored 65.2% vs google's official 47.8%, and the existing top closed source model Junie CLI's 64.3%. Since there are a lot of reports of deliberate cheating on TerminalBench 2.0 lately (https://debugml.github.io/cheating-agents/), I would like to also clarify a few things 1. Absolutely no {agents/skills}.md files were inserted at any point. No cheating mechanisms whatsoever 2. The cli agent was run in leaderboard compliant way (no modification of resources or timeouts) 3. The full terminal bench run was done using the fully open source version of the agent, no…
Apr 2026 · github.com
- 3
- 4
13,000+ MCP servers, skills & plugins for AI coding agents
Jul 2026 · codexmarketplaces.com
- 5MA
This weekend I built a multi-agent coding system which, quite unexpectedly, beat Claude Code on Stanford's Terminal Bench! The architecture is straightforward, consisting of an orchestrator agent that deploys explorer & coder subagents to complete complex terminal based tasks, utilising an intelligent context sharing mechanism along the way which makes it all work. The repo has a lot of technical details, and all the code and prompts for you to play around with if you'd like! I had a lot of fun making this, I hope you have fun reading the README, using it yourself, or even extending it! As…
2025 · github.com
- 6IN
Tl;dr: I trained a classifier to route to the least expensive model and reasoning depth to complete the request. Coupling that with additional automated token efficiency techniques has yielded 3x usage for the same spend. For anyone interested in trying it themselves: https://nerfguard.com Various teammates and I switched over to Codex from Claude Code recently. We still bounce between the tools, but Codex’s speed and steerability coupled with performance gains were hard to ignore. One of the downsides was that the per token pricing kicked in way sooner. This is happening across…
Jun 2026
- 7
- 8OA
We were both genuinely impressed by Claude Code after it helped each of us fix nasty CI problems overnight. Doing those fixes manually would have taken days. After that experience, we each found ourselves struggling through Ctrl+Tab through multiple Claude Code windows in our terminals. While we enjoyed having agents working for us in parallel, context switching and cycling through each terminal tab was a real pain. So we thought: Can we design a TUI dashboard that manages a large swarm of agents in one place? Even better, can agents manage agents hierarchically, like how companies work?…
May 2026 · omar.tech
- 9
- 10RB
We built HALO (Hierarchal Agent Loop Optimizer), an open-source tool for debugging and optimizing AI agents using their execution traces. It’s a loop. Run your agent, feed the traces to HALO, get the report, apply the fixes, then re-run your agent. HALO takes in OTEL compliant traces from AI agents using tracing frameworks such as Langfuse, Arize/OpenInference, or even just plain JSONL. It uses an RLM (Recursive Language Model) to more efficiently break trace analysis into smaller subproblems in order to find recurring patterns across large amounts of data and fix systemic issues that…
Jun 2026 · github.com
- 11
The smallest async RL trainer I could write: one loop that runs REINFORCE on CartPole on a laptop and async GRPO on a cluster (e.g. 8xH100 trainer, 8 vLLM workers, ran as a [SkyPilot job group](https://docs.skypilot.ai/en/latest/examples/job-groups.html) on k8s ). All without Ray or TRL or DeepSpeed etc., workers talk to the trainer over stdlib HTTP.
24d ago · github.com
- 12SR
Hello all, I'm a software developer. Over the last few months more and more of my work has turned into using coding agents instead of typing the whole code myself. Usually a few claude sessions at once, sometimes codex, one per feature or per revealed bug. I ran them in a split terminal for a few weeks, and quickly spotted two main problems. The first is that I couldn't easily tell which agent was stuck waiting on me and which was still working, so I'd cycle through sessions and checking on them. The second one: agents sharing a single branch step on each other. Two of them could be editing…
Jul 2026 · shikigami.dev
- 13MD
We’re excited to share ML-Dev-Bench, a new open-source benchmark that tests AI agents on real-world ML development tasks. Unlike typical coding challenges or Kaggle-style competitions, our benchmark simulates end-to-end ML workflows including: - Dataset handling and preprocessing - Debugging model and code failures - Implementing new model architectures - Fine-tuning and improving existing models With 30 diverse tasks, ML-Dev-Bench evaluates agents across critical stages of ML development. To complement this, we built Calipers, a framework that provides systematic performance evaluation and…
2025 · github.com
- 14MA
We built meta-agent: an open-source library that automatically and continuously improves agent harnesses from production traces. Point it at an existing agent, a stream of unlabeled production traces, and a small labeled holdout set. An LLM judge scores unlabeled production traces as they stream. A proposer reads failed traces and writes one targeted harness update at a time, such as changes to prompts, hooks, tools, or subagents. The update is kept only if it improves holdout accuracy. On tau-bench v3 airline, meta-agent improved holdout accuracy from 67% to 87%. We open-sourced meta-agent.…
Apr 2026 · github.com
- 15AL
Hi HN, I built agentspace because I kept seeing tmux recommended for keeping Claude Code sessions alive over SSH. I find multiplexers painful because they subtly change shell behavior in ways I always forget. They also do not offer a standardized way to run highly permissive agents in a safe way. I also tried worktrees, but it just feels awkward to manage work in that way. Docker can help with all of this, so I thought up a workflow and used some shell aliases to automate spawning Claude/Codex agents into docker containers. I ended up creating a cli to wrap all these methods and would…
Jun 2026 · github.com
- 16CF
Hi everyone! Hope you had a great day so far, and maybe its about to get just a little bit better (thanks Winter ;) So I had way to many terminal windows flying about when using Claude, and kept losing track of which terminal / session / project im in right now. So I built a solution for that, presented it to my team, and now we're using our new tool all the time, and developing it to be the most helpful agents organization tool we can think of. Unfortunatelly, Claude Code only for now, will do Codex soon tho. What it does for you: Main features: - Overview of your projects, with…
Aug 2026 · episko.dev
- 17CB
I built a small benchmark to test CLI coding agents on blind bug detection. A challenger agent injects bugs and writes ground truth (`bugs.json`). A different reviewer agent audits the repo without seeing ground truth, and an LLM matcher scores bug-to-finding assignments. Current run: 50 repos, 150 challenges, 450 reviews, 2,603 injected bugs. Weighted detection: Claude 58.05%, Codex 37.84%, Gemini 27.81%. LLM-judge benchmarks are easy to get wrong, so I’d really appreciate critical feedback on benchmark fairness, scoring/matching methodology, and obvious failure modes I’m missing. Full…
Feb 2026 · github.com
- 181D
We just open-sourced the internal system we built at Assembled for running coding agents as a team. Coding agents worked well for individual engineers, but the surrounding workflow was a bit of a mess. We generally found that many engineers had different MCP connections and context for their agents, personal automations running that other people couldn’t access, and very little introspection for what a human’s input into the coding agent looked like. So we built an internal system that converted coding agents into shared team infrastructure. The system runs Codex, Claude Code, OpenCode, and…
Jun 2026
- 19IS
Hi HN, I built AgenTank. It is a small game where an AI agent writes the logic for your tank. You watch it fight, give strategic feedback, let the agent update the tank code, and send it back into battle. I have run 1,000+ battles on my own tank and spent about $200 in Claude credits improving it. The part I enjoy most is not just winning, but watching the tank make visible mistakes, thinking of a better strategy, and seeing whether Claude can turn that into better code.
May 2026 · agentank.ai
- 20TW
I want to share a new dataset of 331 reward-hackable environments. These are real environments used in Terminal Bench and adjacent benchmarks. I first got interested in this because, as a reviewer of Terminal Bench, I noticed a lot of our tasks were hackable. I also noticed that many contributors to the benchmark do so because it provides credibility when selling environments to labs. Hence, TBench tasks are, in my opinion, held to a higher quality standard than those being used today for RL. No one is spending hours manually reviewing the $1B in tasks being purchased by major labs. As far…
Apr 2026 · github.com
- 21BY
we had hundreds of discussions with engineering leaders over the past few months, and everyone's trying to understand where they are in the AI journey. we collected all this data into a benchmark and built a free grader to let you know where you stand. you answer on a 1–5 scale (e.g., autonomy runs from "suggestions only" to "agents own multi-hour workflows across code, infra, and external systems") - takes about 5 minutes. https://agent-benchmarks.com/software-factory/ waiting for your results!
Jul 2026 · agent-benchmarks.com
- 22RA
Hi HN folks, I have been building AI agents for quite some time now. The shift has gone from LLM + Tools → LLM Workflows → Agent + Tools + Memory, and now we are finally seeing true agency emerge: agents as systems composed of tools, command-line access, fine-grained system capabilities, and memory. This way of building agents is powerful, and I believe it is here to stay. But the real question is: are the systems powering these agents ready for that future? I do not think so. Using Docker for a single agent is not going to scale well, because agents need to be lightweight and fast. LLMs…
Mar 2026 · github.com
- 23
- 24OS
Hello HN, I’ve been building AI agents lately and ran into a common "Context Bloat" problem. When an agent has 20+ skills, stuffing every system prompt, reference doc, and tool definition into a single request quickly hits token limits and degrades model performance (the "lost in the middle" problem). To solve this, I built OpenSkills, an open-source SDK that implements a Progressive Disclosure Architecture for agent skills. The Core Concept: Instead of loading everything upfront, OpenSkills splits a skill into three layers: Layer 1 (Metadata): Light-weight tags and triggers (always loaded…
Jan 2026
Ranked by how close each launch is in meaning, then by votes. Refine with a description →