2026-05-27-weekly-digest-2026-05-13_2026-05-27-frontier-v0

mini-swe-agent is our ~150 line of code agent, and it performs as well as Claude Code, Codex and Gemini CLI on this new coding benchmark.

Yes, I gave it access for Computer Use, not Chronicle, the feature that caused it to record my screen on a regular basis and generate markdown files for additional connect even when I wasn’t using codex

if mini-swe-agent at 150 lines matches the frontier on this benchmark, the benchmark is doing a lot of the work. most production complexity lives in the long tail, not the canonical task shape

DeepSWE finds that mini-swe-agent significantly outperforms ClaudeCode and Codex on the benchmark. The simpler the system, the better it generalizes (and mini's core agent class is just ~150 lines of code)

@NousResearch @Teknium Hermes Agent has a bloat/noise issue. 90+ skills enabled by default. Agents still load all of them even if they are disabled. There is no "enabled" option in config, so even if "disabled" option worked correctly, you'd need to keep updating/maintaining the list as more skills are added to Hermes core. Anything that is added to context that is not needed by an agent is noise. bloat is the ever-increasing skill list and features additions to core. This is important for builders that understand nuance. Any efforts at addressing these?

@robdubparker @NousResearch Yes there is. Use ‘hermes skills config’ and disable whatever you want. This is not bloat though. All 90 skills use about 1.5-2k tokens. Thats less tokens for all skills then just the browser use toolset

Boris Cherny: "Claude Code is 100% written by Claude Code, Cowork is 100% written by Claude Code" across all of Anthropic, 90% of code is written by Claude lawyers, designers, finance people are building real things with Claude Code right now with no CS degree, no background, nothing the people who built the tool use the tool to build the tool think about what that means for everyone else the guide on everything Claude can actually do is in the article below

We use both together. In practice, for containers to be useful, you often have to punch some holes: GitHub, Anthropic API, kube, other MCPs. Auto mode makes interacting with these safer. It significantly reduces the risks of accidental data deletion, exfiltration, and prompt injection.

Just found out that Berkeley course staff are writing hooks inside course repos so if a student opens an assignment in Claude Code or Cursor the agent will automatically ping the staff 😵‍💫 well played

@geminicli Very excited to unify the experience for developers building with Gemini under Antigravity! Hoping this adds more clarity for those asking what product is right for them.

So google is replacing gemini-cli with agy (antigravity cli), but: 1. agy is not opensource 2. It no longer supports ACP Really unfortunate development

Transitioning Gemini CLI users to Antigravity CLI We are unifying our efforts around a single harness and platform, Google Antigravity with four distinct surfaces: • Antigravity 2.0 • Antigravity CLI • Antigravity SDK • Antigravity IDE This will allow us to move faster and give you a streamlined experience wherever you do your best work. Rebuilt in Go for speed, Antigravity CLI is available today and brings robust multi-agent orchestration and asynchronous workflows to your terminal. Important things to know: 1. If you are using Gemini CLI through your Google one account (Google AI Pro or AI Ultra) or through Gemini Code Assist for individuals (free offering) we will be helping you migrate your workflows over the next 30 days. 2. No action required for Enterprise users. Enterprise plans and API keys will continue to be supported in Gemini CLI. Read the full details in our blog post → https://goo.gle/4eWkUgK

5.5 xhigh with the omp /goal mode feels like the first agent i actually trust to leave alone 24/7. codex does not hold a candle

Claude Codeの/configからdefault permissionでauto modeが実装されてめっちゃ便利！ みたいなポスト流れてくるから試しました。 元々、bypassPermissionsでしたがそれよりも確認や判断、決断や操作がめちゃくちゃ増えたので戻しました。

Congratulations to everyone who has discovered /goal in the last two weeks. Copilot CLI shipped `autopilot` mode in January.