The Handoff Has an API

Hermes Agent now has two ways to update: 1. Stable: The default, your Hermes will only update to stable releases 2. Main: The previous approach, your Hermes will update to the very latest commit I made a short video showing the difference, and how to switch between the modes!

what does it tell you that a shitty little coding agent with 4 tools and no subagents is at the pareto frontier?

Origin, Cursor’s code hosting platform, now supports over 20 integrations. Connect your repos to tools for CI, deployments, code review, security, monitoring, and project management.

Sorted this out over DMs but replying here as well in case anyone else hits this! Root cause was PreToolUse hooks returning "allow" by default, which tells Claude that the command is already approved, so it skips the step where the auto mode classifier would normally run. Easy to miss, we're looking at changing this on our end too. In the meantime; if you have any deny hooks, make sure they print nothing instead of "allow" when there's nothing to block! That way the command still goes through the classifier. If you want Claude to always ask before deleting files (even in auto mode) add "Bash(rm *)" under permissions.ask

Codex CLI leaves a self-updating background daemon running after you close the terminal. It is on by default. What is on the record (openai/codex issue #48195, open since Sep 25, on the HN front page today): Since 0.157.0, opening an interactive codex session starts "codex app-server --managed-daemon". It runs with no TTY and gets reparented to PID 1, so it survives every window you close until you stop it or reboot. It copies the whole platform package, about 374 MB including bundled bwrap, zsh and rg, into ~/.codex/packages/app-server-daemon/ and runs from that copy. The backend version is now decoupled from whatever npm or brew installed. A detached updater (pid-update-loop) checks for releases 5 minutes after start, then hourly, downloads them and restarts the backend, which can kill active turns. It also pins the environment of whichever session started it. The reporter's per-invocation HTTPS_PROXY stuck to the daemon's outbound traffic for every later session. And "codex app-server daemon stop" leaves the updater alive. The 0.162 release notes (Oct 8) still tune daemon auto-start rather than drop it. If you run Codex from @OpenAIDevs on a work laptop or CI image, you now have a coding agent that executes commands on your machine and ships its own long-lived, self-replacing binary outside your package manager. Your inventory says one version; the process answering turns may be another. Tonight: 1. Run "codex features disable daemon_auto_start" (or set daemon_auto_start = false in config.toml) on every dev box and CI image 2. Stop it fully: "codex app-server daemon stop", then pkill -f 'app-server daemon pid-update-loop', then delete ~/.codex/packages/app-server-daemon 3. Add a fleet or EDR check for codex processes parented to PID 1 and binaries running from ~/.codex/packages Closing the window should end the agent. https://t.co/lzOCb1sJQP

works until the bug takes down the agent itself. we host openclaw agents and what breaks is rarely the model, its the box: ram blows up mid build, disk fills, logins expire. if nothing outside brings it back, theres no one left to read the error

The loop never closes because "find bugs" has no stop rule, so every fix earns another pass. Done has to come from something binary, a failing test or an unmet spec, not from a reviewer running out of ideas.

Asking Claude to review your code is a slot machine. Every review finds brand new bugs. Review again: more. Review a tenth time: it still finds something. It never runs out of bugs to find... so how do you decide your code is actually done?

2/3 En Copilot CLI: /sandbox enable → activa el aislamiento. /sandbox status → comprueba si se aplica en esa sesión. Ojo: el acceso saliente a internet está permitido por defecto. Activarlo no equivale a desconectar al agente.

Day 2.1/ We have made Auto-review free for all users signed in through a ChatGPT account. You can enable it in settings > permissions > auto-review. Auto-review improves upon the default sandbox setting that requires you to approve everything, which is prone to decision fatigue unless you spend a lot of time configuring specific rules. It allows you to run long tasks while having a second agent review all actions taken by the primary agent. Its only goal is to prevent high-risk actions from being taken and to protect against unwanted actions that are not aligned with the original user intent. This Auto-review feature is now free and does not draw usage from your plan.

Checkpointing is the part we'd test hardest: a resumed run that replays a send is worse than a crash. Our browser job was audited five times by a second model before it ran unattended. When Pi resumes, does it replay side-effecting steps?

before @eve, we had built an agent called fluffles. it ran on a mac mini that sits across me. the goal was for it to be a god agent - an agent so powerful it could operate indistinguishably from a human, with no restrictions whatsoever. we wanted to give it a credit card, phone number, patched browser - the whole 9 yards. we mapped out what primitives it would need, and came up with the architecture shown. we thought it would be straightforward. but building the harness was unbelievably painful. from automations to subagents. there wasn't one specific thing that was hard, it was that the surface area of shit that can go wrong is unbounded. when eve came out, we ported over almost immediately. the scaffolding let us focus on agent behavior instead of the intricacies of a slack integration. we're excited to show you our work, and will be launching this week.

The useful part is the feedback loop, not the one-click fix. Keep the failing input, offending commit, and post-fix regression test attached to the agent task, or you will automate guesswork.

Couldn't have said it better myself. And durability shouldn't be the secondary afterthought bolted on later. Won't be long until we get the Temporal Agent Harness on that shortlist.

When Rollouts catches a regression, it finds the offending PR and opens an issue. One click starts a cloud agent to fix it. Rollouts usage credits are included through Oct 3.

This is so true! At OpenHands we have have reached the "bonus stage" from the original post. Our process has been rebuilt around the assumption that if an issue is opened on the repo, we are guaranteed get a vibe-coded PR opened, so our defense is: (1) make sure that all the issues are maximally clear through automated issue triage, and (2) automate code review so that we only accept high-quality contributions. All of these are powered by OpenHands automations: https://t.co/NW7D3ULZDt

Keep mods to yourself, or submit them to the Claude directory if others might find them useful. Mods run with the same access to your machine as Claude Code itself, so make sure to only install mods from sources you trust. https://t.co/BwAvIVj8Dv

You can now mod Claude Code: - Change how it behaves - Customize the UI - Swap in your own features Write one with a few lines of TypeScript, or have Claude build it for you. Mods ship inside plugins, so you install them with /plugin in the CLI or desktop app. A few examples:

Glad that codemode is disabled by default. I've always had trouble with it (tried in dsh / prime etc). Models keep fighting their own batching issues (scripting problems, or batching incorrectly and then having to re batch again because one command was wrong etc).

A pretty big change is that Pi now has MCP at the core. How is this even possible? What made us change our way? https://t.co/q7M4oM49L1

eve is one of the few frameworks that gives you full control over building the exact agent loop that you want. defaultTools: false go build the exact loop for your use case. all of our default tools are build on public eve api. we give you full access to the background task model, dynamic workflows, parking, HITL. I'm not sure of any other framework that gives you the low level tools that eve gives you.

lots feedback here, many of you are planning yourself & don't need plan mode others prefer the UX of entering a mode where Claude is just thinking & brainstorming with you my plan is to: - make plan mode into a built-in mod - allow mods to add new modes or override shift+tab

we’re thinking of killing plan mode and using the shift+tab hotkey to adjust effort levels I don’t think the models need plan mode anymore, but if you’re a plan mode diehard would love to get your feedback on why

plan mode isn't for the agent it's for me lol