Hacker News

I've been quietly building this forlast few months. Today I am open-sourcing it.

One thing that kept bothering me is that teams I worked with were sending raw user input to OpenAI with literally zero filtering. Not even a regex. Users would type their SSN or credit card number, it would go straight to the API, and nobody noticed. There's no safety net here. The model doesn't care. The SDK doesn't care. Your app doesn't care.

So I built Sentinel Protocol. It's a local proxy that sits between your app and any LLM API - OpenAI, Anthropic, Google Gemini, Ollama,etc and enforces security on every request.

What it actually does: On the way in: - Scans for 40+ PII types (SSN, credit card, email, phone, passport, tax ID, AWS keys, API tokens, etc.), blocks critical ones, silently redacts medium ones - Neural injection classifier (built a custom LFRL engine - rule language plus ML scoring) + regex + semantic similarity - layered defense - MCP poisoning detection for agentic apps using tool calls - Loop detection, intent drift tracking, swarm isolation for multi-agent systems - Deception engine that intentionally returns fake responses to detected attackers - Cold start analyzer (heightened sensitivity during first N seconds of warmup)

On the way out: - Output classifier for toxicity, code execution, hallucination signals, unauthorized disclosure - Hallucination tripwire (catches fabricated URLs, nonexistent citations, numeric contradictions in the model's own response) - Real-time PII redaction in SSE/streaming responses — not after the stream, during - Stego exfil detection (zero-width characters, invisible Unicode used to embed data in model output - real attack vector) - Token watermarking with timing-safe verification

Governance: - OWASP LLM Top 10 - all 10 categories covered - MITRE ATLAS threat attribution on every blocked event - JSONL audit log at ~/.sentinel/audit.jsonl (grep-friendly, plain text, yours) - Forensic debugger with full replay capability — change a config, re-run any blocked request against the new settings - AIBOM (AI Bill of Materials) generator for compliance - TLA+ and Alloy formal verification specs included

Numbers: - 52,069 lines of source code - 81 security engines - 139 test suites, 567 tests, 0 failures - 306 linted files, 0 warnings - 9 total runtime dependencies (yes, nine — I kept it tight on purpose) - <5ms p95 proxy overhead - Zero cloud dependency — everything runs on your machine

Start with one command: npx --yes --package sentinel-protocol sentinel bootstrap --profile paranoid --mode enforce --dashboard

Drop-in for any OpenAI SDK — change baseURL to http://127.0.0.1:8787/v1 and add the x-sentinel-target header. That's it.

I wanted to build something that could run in a hospital, a law firm, or a two-person startup with the same trust model: your data doesn't leave your machine.

GitHub: https://github.com/myProjectsRavi/sentinel-protocol npm: sentinel-protocol (v1.2.7, MIT)

Expecting feedback from every developer.

Comments URL: https://news.ycombinator.com/item?id=47167296

Points: 1

# Comments: 0

Hey HN! I was very early adopter of Arc. as a dev, I enjoyed useful shortcuts (like copy URL), vertical sidebar, command panel, and other cool features. when TheBrowserCompany left Arc, i also switched browsers but missed all those cool features. so I decided to make Arc’s features universal.

Meet SupaSidebar - Arc-like sidebar for all browsers

Here’s what the app can do: - Save links, files, and folders from any browser or Finder - Fuzzy search current open tabs, browser history, and saved links - Open saved links in any browser with just a click - Common browser history across browsers - iCloud sync - Link browser profiles to Spaces and quickly switch between them - Built-in sidebar feel with fullscreen, overlay protection, auto show/hide etc. - Global shortcuts, Air Traffic Control, AI tagging, and more

here's a demo video: [SupaSidebar Demo](https://youtu.be/OimIUO0PSAk)

Free to use for up to 3 Spaces. Native Swift/AppKit, no Electron.

Still in development and I’m making it better daily, so any feedback would be appreciated.

Comments URL: https://news.ycombinator.com/item?id=47167284

Points: 1

# Comments: 0

Article URL: https://arxiv.org/abs/2602.12361

Comments URL: https://news.ycombinator.com/item?id=47167276

Points: 1

# Comments: 0

Article URL: https://www.thegitcity.com/

Comments URL: https://news.ycombinator.com/item?id=47167275

Points: 1

# Comments: 0

I built termstage because I was tired of screen-recording CLI demos and ending up with heavy GIFs that look bad in READMEs.

termstage lets you describe a terminal session in YAML and renders it as an animated SVG — lightweight, crisp at any size, embeds directly in GitHub READMEs.

Example:

steps: - cmd: "notes list" output: "#1 Fix the login bug" - cmd: "notes done 1" output: "Marked #1 as done" Run `termstage render demo.yaml --animated` and you get an SVG you can drop anywhere.

Install: pipx install termstage

GitHub: https://github.com/saikatkumardey/termstage PyPI: https://pypi.org/project/termstage/

Comments URL: https://news.ycombinator.com/item?id=47167274

Points: 1

# Comments: 1

I built an MCP server that gives Claude Code a "phone a friend" lifeline. Instead of relying on one model's perspective, Claude can pull in GPT, Gemini, DeepSeek, or any OpenAI-compatible model for a structured multi-round debate — and participate as an active debater itself.

How it works:

You ask Claude to brainstorm a topic All configured models respond in parallel (Round 1) Claude reads their responses and pushes back with its own take Models see each other's responses and refine across rounds A synthesizer produces the final consolidated output Claude isn't just orchestrating — it has full conversation context, so it knows what you're working on and argues its position alongside the other models. They genuinely build on and challenge each other's ideas.

A 3-round debate with 3 models costs ~$0.02-0.05. One model failing doesn't kill the debate — results are resilient.

npm: npx brainstorm-mcp GitHub: https://github.com/spranab/brainstorm-mcp Sample debate (GPT-5.2 vs DeepSeek vs Claude): https://gist.github.com/spranab/c1770d0bfdff409c33cc9f985043...

Free, MIT licensed. Works with any OpenAI-compatible API including local Ollama.

Comments URL: https://news.ycombinator.com/item?id=47167271

Points: 1

# Comments: 0

Article URL: https://home.phyz.dev

Comments URL: https://news.ycombinator.com/item?id=47167270

Points: 1

# Comments: 0

The current binary computing paradigm has hit a hard thermodynamic wall. We are watching the most advanced digital systems throttle their own logic because the energy cost of algorithmic emulation, forcing continuous reality through billions of discrete, microscopic gates, has become fundamentally unsustainable. For the past decade, I have been engineering the foundation of a new era: a non-algorithmic architecture that utilizes 3-phase AC power directly as the computational medium. Instead of binary voltage thresholds, the architecture computes using the continuous phase differential of macroscopic wave superposition. The foundational logic is driven by the physical integration of phase vectors:

V_net(t) = Σ [A_n * sin(ωt + Φ_n)]

By engineering the physical medium to capture specific interference patterns, we achieve true ternary logic states based on thermodynamic phase tension, natively undercutting the Landauer limit of binary bit-erasure:

[ 1 if ΔΦ > θ_th ] L(ΔΦ) = [ 0 if |ΔΦ| ≤ θ_th ] [ -1 if ΔΦ < -θ_th ]

Because there is zero algorithmic translation, the continuous wave translates instantaneously into discrete natural language tokens. It doesn't just calculate; it generates complex vector embeddings and fluid natural speech as a single physical event, completely bypassing the massive GPU arrays currently suffocating under LLM power demands. Applications: The implications extend far beyond the datacenters. In orbital space missions, where cosmic radiation regularly causes catastrophic binary bit-flips, this architecture is completely immune. You cannot bit-flip a continuous AC wave, providing unbreakable physical autonomy for rovers and satellites.

Comments URL: https://news.ycombinator.com/item?id=47167267

Points: 1

# Comments: 0

Article URL: https://www.extruct.ai/research/industrial-tech-thesis/

Comments URL: https://news.ycombinator.com/item?id=47167263

Points: 1

# Comments: 0

Article URL: https://campedersen.com/phyz-at-home

Comments URL: https://news.ycombinator.com/item?id=47167258

Points: 1

# Comments: 0

Article URL: https://garden-horizons.com/

Comments URL: https://news.ycombinator.com/item?id=47167250

Points: 1

# Comments: 0

Article URL: https://www.propublica.org/article/dilley-detention-center-kids-art-removal

Comments URL: https://news.ycombinator.com/item?id=47167248

Points: 3

# Comments: 2

Article URL: https://github.com/letsur-dev/ccperm

Comments URL: https://news.ycombinator.com/item?id=47167242

Points: 1

# Comments: 0

Article URL: https://twitter.com/LindsayCasale/status/2026758489276821655

Comments URL: https://news.ycombinator.com/item?id=47167222

Points: 1

# Comments: 0

Hi HN,

I’m the creator of StageWright (and the open-source playwright-smart-reporter).

I’ve been frustrated by the "black box" nature of E2E test failures. Standard reporters tell you that a test failed, but they don't help you understand why it’s failing across 50 different runs or whether its execution time is trending toward a regression.

I built StageWright to treat test results as a performance and stability dataset.

Key Technical Features:

Historical Flakiness Detection: Unlike Playwright's default "retry" logic, we track failures across runs. A test only gets a high "Stability Grade" if it consistently passes over time.

Flamechart Step Timelines: We added a color-coded flamechart for test steps (v1.0.8). It categorizes steps into Navigation, Action, and API, making it easy to see if a 10s test is hanging on a locator or a slow backend response.

2-Sigma Anomaly Detection: The trends view uses moving averages and 2-sigma outlier detection to flag performance regressions that might otherwise go unnoticed.

AI-Powered Failure Clustering: We batch failures and use Claude/GPT-4 to cluster similar errors. Instead of 20 separate failures, you see "1 cluster: TimeoutError on payment-submit-btn."

Virtual Scroll Performance: We optimized the UI with virtual scrolling to handle suites with 500+ tests without the browser freezing—a common issue with the default HTML reporter.

Native Trace & Network Logs: Traces and network waterfalls are embedded directly in the report. No downloading .zip files from CI; they open instantly in an inline viewer.

The Architecture: StageWright is built to be "Playwright-native." It hooks into the reporter API and can run locally (outputting a standalone HTML/JSON history) or via our new Starter/Pro cloud tiers. The Pro tier provides a centralized dashboard for teams, long-term history retention, and cross-project analytics.

I’m currently supporting both Node.js and Python (pytest-playwright) environments.

I’d love to hear what the community thinks—especially regarding how you handle "test debt" in large CI pipelines. I'm here for any questions!

Comments URL: https://news.ycombinator.com/item?id=47167217

Points: 1

# Comments: 0

Article URL: https://arstechnica.com/space/2026/02/a-non-public-document-reveals-that-science-may-not-be-prioritized-on-next-mars-mission/

Comments URL: https://news.ycombinator.com/item?id=47167216

Points: 1

# Comments: 0

Article URL: https://ia.net/topics/trapped-in-ms-office

Comments URL: https://news.ycombinator.com/item?id=47167214

Points: 1

# Comments: 0

Article URL: https://github.com/xeloxa/WP-Hunter

Comments URL: https://news.ycombinator.com/item?id=47167211

Points: 1

# Comments: 0

Article URL: https://www.erininthemorning.com/p/kansas-sends-letters-to-trans-people

Comments URL: https://news.ycombinator.com/item?id=47166522

Points: 1

# Comments: 0

Article URL: https://www.youtube.com/watch?v=aoag03mSuXQ

Comments URL: https://news.ycombinator.com/item?id=47166473

Points: 1

# Comments: 0