frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Berkshire Has a Website from the '90s and Buffett Fans Say Don't Mess with It

https://www.wsj.com/tech/personal-tech/berkshire-hathaway-shareholder-meeting-warren-buffett-greg...
1•firexcy•45s ago•0 comments

How the Legal Opium Market Shaped Global Trade–and Led to an Opioid Crisis

https://www.bu.edu/articles/2026/how-the-legal-opium-market-led-to-an-opioid-crisis/
1•hhs•1m ago•0 comments

Former head of 'Pentagon's think tank' joins Anthropic

https://www.defenseone.com/technology/2026/05/former-head-pentagons-think-tank-joins-anthropic/41...
1•Jimmc414•4m ago•0 comments

Tesla owner won $10k in court for Tesla's FSD lies. Tesla is still fighting him

https://electrek.co/2026/05/02/this-tesla-owner-won-10k-in-court-for-teslas-fsd-lies-tesla-is-sti...
1•breve•5m ago•0 comments

Show HN: Language app with spaced repetition and comprehensible input

1•ChadNauseam•5m ago•0 comments

The Claude Delusion: Richard Dawkins believes his AI chatbot is conscious

https://www.dailygrail.com/2026/05/the-claude-delusion-richard-dawkins-believes-his-female-ai-cha...
1•SwellJoe•6m ago•0 comments

Google Summer of Code 2026 selected projects

https://blog.rust-lang.org/2026/04/30/gsoc-2026-selected-projects/
1•kazu11max17•8m ago•0 comments

AI agents are briefly overhyped

https://stevekrouse.com/agent-hype
1•stevekrouse•18m ago•0 comments

To Make Orchestras More Diverse, End Blind Auditions

https://www.nytimes.com/2020/07/16/arts/music/blind-auditions-orchestras-race.html
1•bilsbie•20m ago•0 comments

Meta faces New Mexico trial that could force change to Facebook, other platforms

https://www.reuters.com/legal/government/meta-faces-new-mexico-trial-that-could-force-changes-fac...
3•1659447091•29m ago•0 comments

The Race Is on to Find the Treasure Buried in San Francisco

https://www.nytimes.com/2026/05/02/us/san-francisco-buried-treasure-chest.html
1•mistersquid•32m ago•0 comments

AWS Lightsail's $0.09/GB Bandwidth Overage Is a Trap for Small Projects

https://galaxycloudsolutions.com/blog/aws-lightsail-vs-galaxy-cloud-solutions/
2•rougereaper420•34m ago•0 comments

With $1 Cyberattacks on the Rise, Durable Defenses Pay Off

https://spectrum.ieee.org/ai-cyberattacks-memory-safe-code
1•rbanffy•41m ago•0 comments

Coatue has a plan to buy up land for data centers, possibly for Anthropic

https://techcrunch.com/2026/05/01/coatue-has-a-plan-to-buy-up-land-for-data-centers-possibly-for-...
1•Brajeshwar•41m ago•0 comments

The Computer Programme Episode 1, 1982 [video]

https://archive.org/details/the_computer_programme_ep01
2•petethomas•42m ago•0 comments

Voice-AI-for-Beginners – A curated learning path for developers

https://github.com/mahimairaja/voiceai
2•mahimai•47m ago•0 comments

Restorative Yoga and the Biology of Belonging

https://parrik.com/puzzles/the-partition-problem/
1•parrik•47m ago•0 comments

Facepunch launches s&box, the highly anticipated successor to Garry's Mod

https://www.gamingonlinux.com/2026/04/facepunch-launches-s-box-the-highly-anticipated-successor-t...
5•embedding-shape•49m ago•1 comments

Dynamic Traefik configuration with multiple Docker hosts

https://blog.vasi.li/automating-mantrae-traefik-management-with-mantrae-agent/
2•vsviridov•50m ago•0 comments

Grinta – Local-first coding agent, 7 months solo, open source today

https://github.com/josephsenior/Grinta-Coding-Agent
1•YoussefMejdi•51m ago•1 comments

Trump's border wall expansion just bulldozed an ancient tribal site

https://www.washingtonpost.com/climate-environment/2026/04/30/border-wall-damage-indigenous-arizona/
5•gnabgib•53m ago•0 comments

What Is GStack? Gary Tan's Open-Source Startup Framework for Claude Code

https://www.mindstudio.ai/blog/what-is-gstack-gary-tan-claude-code-framework
2•evo_9•56m ago•0 comments

The physics slop that YouTube wants me to make [video]

https://www.youtube.com/watch?v=Cd5EHfRerGI
2•surprisetalk•59m ago•0 comments

Built this for my civil engineering firm's static site on Cloudflare Pages

https://github.com/bwengr/knowledge-base-spec
1•bwengr•59m ago•0 comments

How to run a cross-cutting campaign

https://parrik.com/puzzles/the-campaign-cascade/
1•parrik•1h ago•0 comments

NovAST

https://github.com/sharkkyyy10/NovAST
3•sharkkyyy10•1h ago•0 comments

The Apprehension Engine (2022)

https://guitar.com/features/interviews/the-apprehension-engine-most-terrifying-musical-instrument/
1•turtleyacht•1h ago•1 comments

A self was never flat

https://parrik.com/puzzles/know-thyself/
1•parrik•1h ago•0 comments

Martian Glaciers with Drones

https://nautil.us/uncovering-hidden-martian-glaciers-with-drones-1280400
1•Brajeshwar•1h ago•0 comments

talkie-coder: From 1930 to SWE-bench

https://github.com/RicardoDominguez/talkie-coder
2•Philpax•1h ago•0 comments
Open in hackernews

The agent harness belongs outside the sandbox

https://www.mendral.com/blog/agent-harness-belongs-outside-sandbox
26•shad42•1h ago

Comments

Retr0id•50m ago
It took me a while to grok why this made any sense, I think the context is that this is for hosting many agents as a service.
qezz•45m ago
Exactly, my understanding is also that they host agents as a service. The actual use case is mentioned in the end of the article, which makes it hard to reason about.

Anyway. General advice: treat harnesses as any other (third-party) software that you run on your server. Modern harnesses (the ones from big companies, you need to subscribe to) are black boxes. Would you run a random binary you fetched from the internet on your server? Claude code, codex etc. are exactly this.

shad42•32m ago
We don't host 3rd party agents (I don't know if this what you implied). We built an agent that monitors CI pipelines, tests failures, performance and auto opens PR to address issues we find. We host our agent loop on a backend (it's in go), and we call to the sandbox when we run operations involving the user code.
saltcured•50m ago
Sure, the experimental, agentically-developed code should be tested in a sandbox. This sandbox should contain the damage of the code execution when it goes wrong.

But shouldn't there really be another sandbox where the agentic tool calls execute? This is to contain the damage of the tool execution when it goes wrong.

And, the agent harness itself should either implement or be contained in a third sandbox, which should contain the damage of the agent. There should be a firewall layer to limit what tool requests the agent can even make. This is to contain the damage of the agent when it formulates inappropriate requests.

The agent also should not possess credentials, so it cannot leak them to the LLM and allow them to be transformed into other content that might leak out via covert channels.

shad42•39m ago
Yes, it's also because the agent described in the post is doing some operations on the user code (fix CI pipelines, rerun tests, fix them, etc...). So another big reason to use the sandbox is to run things like bash on a user code. you don't want credentials or anything trusted inside that sandbox, including the LLM api key.
aluzzardi•31m ago
Author here. Depending on how it’s designed, the harness itself doesn’t need any sandboxing.

At the end of the day, it’s a “simple” loop that calls an external API (LLM) and receives requests to execute stuff on its behalf.

It’s not the agent running bash commands: you (the harness author) are, and you’re in full control of where and how those commands get executed.

In the article’s case, bash commands are forwarded to a sandbox, nothing ever runs on the harness itself (it physically can’t, local execution is not even implemented in the harness).

trjordan•38m ago
Nah. Worse is better.

The reason agents work is because they have access to stuff by default. The whole world is context engineering at this point, and this proposal is to intermediate the context with a bespoke access layer. I put the bare minimum into getting my dev instance into a state where I can develop, because doing stuff (and these days: getting my agent to do stuff) is the goal.

This makes slightly more sense if you're building a SaaS and trying to get others to give you access to their code, their documents, and the rest so you can run agents against it. But the easiest, most powerful way is to just hook the agents up to the place that's already set up.

ossa-ma•5m ago
They are building exactly what you described and this is their architectural solution to ensuring their YOLO agents do not nuke their customers code/documents/databases by sandboxing everything in the workspace — the git checkout the agent is working on, plus whatever's needed to run commands against it (compilers, package managers, etc.).
8thcross•34m ago
we are running a harness outside the sandbox, inside a sandobx.
blcknight•24m ago
I am not sure anyone knows what a harness is at this point. I've heard 17 different definitions of it at this point. It's almost like a buzzword in search of a problem.
aluzzardi•18m ago
Author here. My definition is: you take an agent, remove the model and you’re left with the harness.

Tools, memories, sandboxing, steering, etc

irishcoffee•9m ago
I don’t even know what an agent means, let alone harness.
solidasparagus•24m ago
Why are two concurrent sessions updating the same memory key with different values? IMO it probably points to a fundamental flaw in how memory is being thought about and built.
aluzzardi•10m ago
Author here. Because of parallelism and non determinism.

This problem is quite common and not limited to memories. For instance, Claude Code will block write attempts and steer the agent to perform a read first (because the file might have been modified in the meantime by the user or another agent).

Same principle here: rather than trying to deterministically “merge” concurrent writes, you fail the last write and let the agent read again and try another write

Koffiepoeder•5m ago
Slightly related: I am looking for:

- Easy single command CLI agent spawning with templates

- Automatic context transfer (i. e. a bit like git worktrees)

- Fully containerised, but remote (a bit like pods)

- Central, mitm-proxy zero trust authn/authz management (no keys or credentials inside the agents), rather enrichment in the hypervisor/encapsulation

- Multi agent follow-up functionalities

- Fully self hosted/FOSS

Basically a very dev-friendly, secure, "kubernetes"-like solution for running remote agents.

Anyone has an idea of how to achieve this or potential technologies?