Audited August 3, 2026 against 185 MDX pages in steel-dev/docs.
Overall: about 6/10 adherence. The docs contain all four Diátaxis modes, and several pages execute them well, but the modes are not consistently separated or clearly signposted. The strongest areas are how-to and reference; explanation and beginner tutorials are underdeveloped.
A playbook + lessons learned from pulling a Keynote deck out of a long YouTube
video. The slides lived in a small picture-in-picture box in the top-right
corner. Goal: get one image per distinct slide, then OCR the deck to
Markdown. We only cared about a 15-minute window (05:05:00 → 05:20:00).
Run: Stealth_Bench_V1 · framework browser-use (installed) · browser steel · model claude-opus-4-8
Started: 2026-06-19 18:39:20 (local) · Duration: ~6.4 h (23,066 s) · Tasks: 80 completed
Headline: 52 pass / 28 fail (65%). ~93% of failures are bot-detection, not infra.
Run data: full per-task traces and the summary JSON are published at https://github.com/steel-experiments/bu-bench-results.git — see
run_data/Stealth_Bench_V1_framework_browser-use_browser_steel_model_claude-opus-4-8_start_at_20260619_183920/(80 task.jsontraces) andresults/Stealth_Bench_V1_framework_browser-use_browser_steel_model_claude-opus-4-8.json.
Deep-research framework where the primary builder is a coding agent, not a human developer.
Status: Design proposal. Not yet implemented. Current @steel-dev/atlas (v0.1.x) ships a single opinionated research() function and CLI. This document specifies how to evolve it into an SDK for building domain-specific researchers, with Agent Experience (AX) as the primary design constraint.
| <!DOCTYPE html> | |
| <html lang="en"> | |
| <head> | |
| <meta charset="utf-8"> | |
| <meta name="viewport" content="width=device-width, initial-scale=1"> | |
| <title>Context Rot — A Cross-Disciplinary Research Review</title> | |
| <style> | |
| :root { | |
| --bg: #fafafa; | |
| --text: #1a1a2e; |
ClawBench runs 153 tasks on live production websites (Uber Eats, Booking.com, Robinhood, etc.) inside local Docker containers with a DIY Chromium stealth layer. Steel solves exactly this problem as a managed service. The integration surface is minimal — browser-use already accepts a custom CDP endpoint, meaning the harness change is one line of code.
ClawBench is also the most credible AI browser agent benchmark published to date (arXiv
A pattern for building personal knowledge bases using LLMs.
This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.
Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.
Date: April 6, 2026
Method: Steel browser session (logged in as nibzard with "showdead" enabled), JS-based DOM extraction
Scope: 23 pages of /newest (~690 stories), 100 domains investigated via /from?site=
Posts marked [dead] have been killed by HN moderators or the anti-spam algorithm. They are invisible to logged-out users and to users who haven't enabled "showdead" in their profile. They are NOT the same as HTTP 404s — the linked URLs may still work fine. It's a content moderation signal.