Feed aggregator

Article URL: https://fd.xyz/finance-district-blog/why-crypto-ux-is-broken-and-how-agents-might-fix-it

Comments URL: https://news.ycombinator.com/item?id=47128754

Points: 1

# Comments: 0

Here are hints and the answers for the NYT Connections: Sports Edition puzzle for Feb. 24, No. 519.

Here are hints and answers for the NYT Strands puzzle for Feb. 24, No. 723.

Here are some hints and the answers for the NYT Connections puzzle for Feb. 24 #989.

Article URL: https://chaos.social/@doener/116121954248163875

Comments URL: https://news.ycombinator.com/item?id=47128745

Points: 1

# Comments: 0

Here are hints and the answer for today's Wordle for Feb. 24, No. 1,711.

See if you qualify for one of these student-focused discounts.

Article URL: https://medium.com/@tom_84912/happy-birthday-xdp-a971b8ac75e6

Comments URL: https://news.ycombinator.com/item?id=47128703

Points: 1

# Comments: 0

Article URL: https://github.com/teamyavio/yavio

Comments URL: https://news.ycombinator.com/item?id=47128702

Points: 1

# Comments: 1

I started to teach myself to code around 15 years ago. At the time I was working service industry jobs (restaurant kitchens, coffee shops) and desperate to change my career and life.

Around that time there was a new thing called available in HTML5 that you could use to render graphics in web pages without plugins; despite my limited knowledge of tech and lack of coding skills, I knew I wanted to be a part of that.

Around that same time, I had also enrolled at the community college for night classes in math. I wanted to "work in tech" after all, so a mathematics brush-up seemed wise. I had always struggled with math. But now that I was both incentivized to learn it, I found myself facing a familiar terror: trigonometry.

The other students, just out of high school, seemed to know this stuff. I was simply confused. The same frustration I had felt in high school returned: No one could clearly answer why this stuff worked. Why did we need a "sine" function?

So I decided to combine my struggles and make them into a single creative project: I would code a way to visualize and understand the trig functions - and in so doing, I would learn trigonometry and HTML5.

I called it "Touch Trigonometry" and the idea was simple: You move your mouse, and the visualization of the trig functions would update. There would be a unit circle showing the angle. There would be a graph showing the plots. As a kinesthetic learner, I theorized that if I could simply "play with trigonometry" I could get an intuitive understanding of trig.

In the end, building it helped me grasp a concept that had always flummoxed me. And to my surprise, it got posted on HN and made it to the front page. It got many thousands of views and lots of comments on its Disqus message board.

Time went by; I moved onto other things; I gave up the domain and forgot about it.

Early this year I decided to see how quickly I could re-implement Touch Trigonometry again, this time in React Native, using Claude Code, in the form of an app. It took me around 8 hours of total coding time (that's including my own pondering, but not stuff like pushing to the App Store). It was easier this time, not just because of the AI agent at my disposal, but also because I already had a foundation in trig, much more experience as a software engineer, and a more concrete sense of what I wanted the UI to be.

There were a lot of friction points with Claude Code (circa Jan 2026). It could not "one-shot" this app, at least nothing like what I really wanted. It made poor recommendations on rendering and UI setup. UI rendering performance was bad at first, especially layering add'l visualizations like labels; only with considerable "coaching" cold Claude make it tolerable.

Notably Claude also had problems comprehending where to place elements in 2D space - especially with the default landscape orientation of the app. I had to manually place elements on the canvas to use as references it could understand. I had to rename many inaccurately named variables by hand - because its own naming conventions were leading it astray and causing more positioning and orientation mistakes. Even despite such interventions, it often did the orthogonal of what I asked.

I would guess that Claude's issues with spatial understanding are easily fixable on Anthropic's side. They could a bajillion synthetic examples to Claude's training data to give it a better understanding of how natural language spatial positioning instructions map to discrete code edits.

That said, once the essence was working, it was pretty simple to go from 90% complete to 99% complete. This app isn't perfect, and it needs a lot more refinement, but it was an interesting exercise to test what Claude Code's limitations are today.

Anyway, Touch Trig is now alive again and I overall it was a super fun side project to work on. It's very satisfying to see my old UI alive again in app form.

Comments URL: https://news.ycombinator.com/item?id=47128665

Points: 1

# Comments: 0

Article URL: https://www.theguardian.com/lifeandstyle/2026/feb/22/a-new-start-after-60-i-baked-a-pie-every-day-for-a-year-and-it-changed-my-life

Comments URL: https://news.ycombinator.com/item?id=47128645

Points: 1

# Comments: 0

Article URL: https://www.colino.net/wordpress/archives/2026/02/23/the-challenges-of-porting-shufflepuck-cafe-to-the-8-bits-apple-ii/

Comments URL: https://news.ycombinator.com/item?id=47128631

Points: 1

# Comments: 0

Claude Code forgets everything between sessions. After months of heavy use, I had 1.6GB of JSONL session files buried in ~/.claude/projects/. Architecture decisions, debugging breakthroughs, and solutions I couldn't find again. There’s nothing more annoying that solving the same problem twice.

I tried RAG pipelines and local vector DBs, but they added complexity for a problem that's really just text search over structured files. So I built search-sessions: a single Rust binary that searches your session history directly. No database, no indexing step, no dependencies.

It supports two modes: index search (~18ms) for finding the right session, and deep search (~280ms with ripgrep, ~1s without) for matching against full message bodies. Each result includes the session UUID so you can resume any conversation with `claude --resume`.

The design choice that makes it work: skip the database entirely. JSONL files are already structured — just search them fast. Under the hood it's just files and a CLI, so it's easy to inspect, backup, and delete. You install it as a Claude Code skill, then ask "do you remember that auth refactor?" and Claude searches your history and picks up the thread. It also works with OpenClaw.

MIT licensed. macOS and Linux:

brew install sinzin91/tap/search-sessions

or

cargo install search-sessions

Curious what people think about exact vs fuzzy vs semantic search as a default — and which other coding CLIs you’d like supported.

Comments URL: https://news.ycombinator.com/item?id=47128630

Points: 2

# Comments: 1

Article URL: https://joplinapp.org/news/20260223-warrant-canary/

Comments URL: https://news.ycombinator.com/item?id=47128628

Points: 1

# Comments: 1

Article URL: https://www.icij.org/news/2026/02/nearly-half-of-powerful-50-caliber-ammo-seized-by-mexican-government-came-from-us-army-plant-defense-minister-says/

Comments URL: https://news.ycombinator.com/item?id=47128604

Points: 1

# Comments: 0

Hey HN,

Most of us want to run autonomous agents like OpenClaw locally, right on our daily-driver machines. We want them to actually interact with our files, our IDEs, and our real local workflows.

But doing that right now is a security nightmare. A hallucinating (or hijacked) agent with raw system access is basically local remote code execution waiting to wipe a directory or leak an SSH key.

The current workarounds are terrible:

The Hardware Air-Gap: Buying a dedicated Mac Mini or a secondary cheap laptop just to run agents safely.

The Cloud: Spinning up a VPS, which adds annoying latency, costs money, and completely kills the "local-first" advantage.

Containers: Shoving the agent in a Docker container. It’s clunky, and it completely lobotomizes the agent because it can no longer see or interact with your actual host OS where your real work lives.

I got fed up and built Raypher.

Raypher is a bare-metal security architecture specifically designed for AI agents. It lets you run agents directly on your host machine while keeping them on a cryptographically tight leash.

How the Alpha works under the hood:

Zero-Touch Interception: You don't have to rewrite your Python/Node agent scripts. Raypher currently uses transparent system proxy injection (netsh portproxy on Windows) to catch all agent traffic automatically.

Hardware Identity: We bind the agent's session cryptographically to your physical TPM 2.0 chip.

The Policy Engine: Every request is evaluated against strict budget, domain, and time-restriction policies. If it tries to phone home to a weird domain, it gets blocked. It also runs local DLP to automatically redact API keys or any sensitive information that you can custom to be redacted before they hit the stream.

We will distribute it as a single script that sets up the interception and launches a local Command Center dashboard so you can watch your agents get secured in real-time.

(Transparency note: The heavy kernel-level eBPF/WFP hooks are currently in development, but this proxy MVP proves the hardware-bound Policy Engine actually works).

join the waitlist/architecture is at raypherlabs.tech to be the first to be notified when we launch

Comments URL: https://news.ycombinator.com/item?id=47128583

Points: 1

# Comments: 0

ExpressVPN is also expanding its reach to virtual reality through support for the Meta Quest platform.

Hey all, I built babyshark, a terminal UI for PCAPs aimed at people who find Wireshark powerful but overwhelming.

The goal is “PCAPs for humans”: Overview dashboard answers what’s happening + what to click next

Domains view (hostnames first) → select a domain → jump straight to relevant flows (works even when DNS is encrypted/cached by using observed IPs from flows)

Weird stuff view surfaces common failure/latency signals (retransmits/out-of-order hints, resets, handshake issues, DNS failures when visible)

From there you can drill down: Flows → Packets → Explain (plain-English hints) / follow stream

Commands: Offline: babyshark --pcap capture.pcap

Live (requires tshark): babyshark --list-ifaces then babyshark --live en0

Repo + v0.1.0 release: https://github.com/vignesh07/babyshark

Would love feedback on UX + what “weird detectors” you’d want next.

Comments URL: https://news.ycombinator.com/item?id=47128535

Points: 2

# Comments: 0

Article URL: https://github.com/openai/openai-python/pull/2874

Comments URL: https://news.ycombinator.com/item?id=47128519

Points: 1

# Comments: 0

Article URL: https://techcrunch.com/2026/02/23/openai-calls-in-the-consultants-for-its-enterprise-push/

Comments URL: https://news.ycombinator.com/item?id=47128518

Points: 1

# Comments: 0