frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Show HN: Meddle – AI-powered IIoT platform for small manufacturers

https://www.meddleconnect.com/en
1•MicheleLacorte•39s ago•0 comments

My custom agent used 87% fewer tokens when I gave it Skills for its MCP tools

https://seroter.com/2026/03/16/my-custom-agent-used-87-fewer-tokens-when-i-gave-it-skills-for-its...
1•tanelpoder•1m ago•0 comments

Why does a Stochastic Parrot make sense at all?

https://causalitylimited.substack.com/p/the-meaning-of-meaning
1•causalityltd•2m ago•0 comments

Capyra – open-source agent runtime for SAP B1 and WhatsApp

https://github.com/marcostaira/capyra
1•marcostaira•3m ago•0 comments

The environmental cost of datacentres is rising. Is it time to quit AI?

https://www.theguardian.com/australia-news/2026/mar/13/ai-datacentres-environmental-impacts
1•voxadam•4m ago•0 comments

A Couple of Git Nits

https://blog.gitbutler.com/git-ux-rant
1•aspleenic•4m ago•0 comments

Are we ready for film distribution via USB drives?

https://www.theguardian.com/film/2026/mar/16/film-distribution-usb-drives-video-store-age
1•n1b0m•6m ago•0 comments

I Take My Laptop to the Gym So Claude Doesn't Have Downtime

https://www.claudecodecamp.com/p/i-take-my-laptop-to-the-gym-so-claude-doesn-t-have-downtime
1•aray07•6m ago•0 comments

Show HN: X07, compiled language where agents write correct code on the first try

https://x07lang.org/
1•webodik•6m ago•0 comments

The 3-Day Starter Plan for Raspberry Pi Beginners

https://raspberrytips.com/beginner-starter-plan/
1•joebig•9m ago•0 comments

Contiguitas: The Pursuit of Physical Memory Contiguity in Datacenters

https://danglingpointers.substack.com/p/contiguitas-the-pursuit-of-physical
1•blakepelton•9m ago•0 comments

Wanted: Europe's Missing Cloud Provider

https://spectrum.ieee.org/europe-cloud-sovereignty
1•Brajeshwar•9m ago•0 comments

Free tool to compare SASE vendors side-by-side

https://sasecompare.com/
1•Shellomo•10m ago•1 comments

Revealed: The worst mega-leaks of methane driving global heating

https://www.theguardian.com/environment/2026/mar/17/revealed-world-worst-methane-leaks-global-hea...
1•guerby•10m ago•1 comments

Death of a Strawman: The Epistemology of a Language Model

https://mvaleadvocate.substack.com/p/death-of-a-strawman-the-epistemology
1•mannykannot•10m ago•0 comments

Ask HN: With Promptfoo acquired by OpenAI, what are MCP devs using for testing?

1•warmcat•11m ago•0 comments

Show HN: Specifica – an open format for writing software specs as Markdown

https://specifica.org/
1•openmason•12m ago•0 comments

Show HN: I'm trying to help aspiring Data Analysts

https://d8a.academy/
3•mariusMDML•13m ago•0 comments

UK security adviser attended US-Iran talks and judged deal was within reach

https://www.theguardian.com/world/2026/mar/17/uk-security-adviser-attended-us-iran-talks-and-judg...
2•prmph•14m ago•0 comments

The Great Developer Schism: Process vs. Product [video]

https://www.youtube.com/watch?v=-zK6JG94pdo
1•Gooblebrai•14m ago•0 comments

Show HN: MCP Isn't Dead. You're Just Using It Wrong

https://www.commandable.ai/blog/mcp-isnt-dead
3•isaacrolandson•14m ago•1 comments

CBM-BASIC: Commodore BASIC–style interpreter written in C

https://github.com/omiq/cbm-basic
1•ingve•17m ago•0 comments

A collaborative pixel mural where each 16×16 tile is owned and editable

https://nftmural.io
1•UPelsin•17m ago•1 comments

X11 user daemon to automatically run commands triggered by user specified events

https://codeberg.org/NRK/xuv
1•PaulHoule•20m ago•0 comments

Nvidia Built the A.I. Era. Now It Has to Defend It

https://buzznews.com/news/2e8391b8-5df1-4d49-a178-df15bd702086
1•buzznewswebsite•20m ago•0 comments

Show HN: MUP – Interactive UI inside LLM chat, so anyone can use agentic AI

https://github.com/Ricky610329/mup
1•Ricky_Tsou•20m ago•2 comments

Samsung to Discontinue Galaxy Z TriFold After Just Three Months

https://www.macrumors.com/2026/03/17/samsung-to-discontinue-galaxy-z-trifold/
2•tosh•22m ago•0 comments

VEO – Open-source content-adaptive video encoding optimizer in Go

https://github.com/terranvigil/veo
1•terranvigil•22m ago•1 comments

Trapped Inside a Self-Driving Car During an Anti-Robot Attack

https://www.nytimes.com/2026/03/17/technology/trapped-inside-a-self-driving-car-during-an-anti-ro...
1•JumpCrisscross•23m ago•0 comments

Java 26 Released

https://mail.openjdk.org/archives/list/jdk-dev@openjdk.org/thread/2MXXXBJKTJXQD25Q4XGGINKYA33T7D5I/
1•mkurz•23m ago•0 comments
Open in hackernews

Show HN: SoulGuard, OS-level identity protection for AI agents

https://github.com/mirascope/soulguard
1•teamdandelion•1h ago

Comments

teamdandelion•1h ago
Hi! I've been making increasing use of OpenClaw agents in my life (on dedicated Mac Minis). I'm impressed by their power and flexibility, a lot of which comes from their capacity to self modify their memories, identity, and config. But that power comes with risks. Anyone who can talk to any of my agents has a vector for privilege escalation, for example by persuading them to update openclaw.json to add untrustworthy channels, to update AGENTS.md, etc. This makes me uncomfortable. Even when sandboxed, digital assistants have access to sensitive information and context.

Sure, you can put "Don't take candy from strangers" in the AGENTS.md, but we really need ways to set security boundaries that are enforced by something outside of the agent itself. SoulGuard sets such boundaries, starting with key files like SOUL.md and openclaw.json. It sets OS level filesystem protections to ensure that protected files are read-only, with a staging process to propose changes. Meaning that your agent can propose changes to openclaw.json, but it physically cannot edit the file unless you approve it.

Rights to approve changes is gated by the human user invoking sudo. (If your agent can sudo then it really has keys to the kingdom; don't do that). SoulGuard also has a daemon that can connect to Discord, so that you review and approve changes from within Discord, rather than needing to ssh in for sudo access. I've also added an openclaw plugin which is unnecessary for security guarantees, but helps the agents learn how to use SoulGuard. (This could use a bit more work, right now agents still may need some prompting to use `soulguard stage` in order to propose changes to protected files.)

I'm dogfooding SoulGuard on my own OpenClaw agents. I'd love to hear if others find it useful. Please do try to break the security model and see if you can find any flaws. I've tried to harden SoulGuard against totally compromised agents, but it's new software and I may have overlooked some attacks.

Here's the GitHub (MIT licensed) https://github.com/mirascope/soulguard

And here's the project site :) https://soulguard.ai/