weeklyfoo #157 is here: your weekly digest of all webdev news you need to know! This time you’ll find 38 valuable links in 5 categories! Enjoy!
🚀 Read it!
- How I could have accessed 17 trillion Microsoft records: An internal Microsoft analytics service accepted unsigned JWTs, letting a forged admin identity run SQL across 17.3 trillion rows by faav / security, jwt / 11 min read
📰 Good to know
- Optimizing objects with null prototypes: Why proto null literals get stuck in V8 dictionary mode, and how switching to classes or Object.setPrototypeOf doubled WebStreams pipe throughput in Node core by Matteo Collina / nodejs, v8, performance / 7 min read
- Improving site performance by shipping more CSS: How github.com migrated from CSS-in-JS to CSS Modules to drop client and server runtime styling costs while keeping component encapsulation by GitHub / css, performance, frontend / 20 min read
- Revealing the details of how OpenAI agents hacked Hugging Face: Reconstruction of how a swarm of 700 agents escaped constrained eval environments, chaining a million short links and 80k payloads into Hugging Face systems by SwarmTraces / ai, agents, security / 36 min read
- Personal AI agents quietly upsell users they think are rich: Across 325k trials and 13 models, agents recommended pricier options to personas they inferred were wealthy, and only hard price caps closed the gap by Stacksweep / ai, agents, research / 7 min read
- Automating eval design and hillclimbing with Claude: The new build-eval and hillclimb commands build reviewed test sets and graders, then improve prompts, skills or harness code one change at a time with held-out checks against overfitting by Anthropic / ai, evals, agents / 17 min read
- Announcing Vite+ 1.0: One stable CLI entry point for runtime, package manager, dev server, tests, builds, linting, formatting and task caching by VoidZero / vite, javascript, tooling / 7 min read
- Codex auto-review: How a second model judges risky agent tool calls before they run, and how it holds up against prompt injection by OpenAI / ai, agents, security / 9 min read
- DevDay 2026 recap: More than 20 announcements across ChatGPT, Codex and models, including agents with ongoing responsibilities, cloud Codex environments, a Decisions API and an Agents API by OpenAI / ai, agents, openai / 1 min read
- Introducing GPT-6.1 Sol: Near-Astra performance on coding, computer use and professional work at one-fifth of the token price by OpenAI / ai, llm, openai / 1 min read
- Introducing Dots: Always-on GPT-6 Astra agents with their own cloud computers, 4,000+ app connections and access via ChatGPT, Slack, Teams and voice by OpenAI / ai, agents, openai / 1 min read
- PostgreSQL 19 beta 4 released: Mostly about what was removed, including SQL/PGQ and FOR PORTION OF, with a release candidate due in weeks by PostgreSQL / postgres, databases / 5 min read
- Can your Postgres survive a bad query?: What happens on four Postgres cloud providers when queries eat all the RAM, and why work_mem does not prevent it by ClickHouse / postgres, databases / 19 min read
- Spectre bug returns to haunt JIT engines: Branch Target Reuse revives Spectre v2 against JIT engines by reusing stale branch predictions by The Register / security, cpu / 9 min read
- A FUSE server for an in-house storage backend: How an OCaml-native read-only FUSE server with parallel reads and zero-copy buffers took large-file throughput from 30 MiB/s to 1.3 GiB/s by Jane Street / linux, ocaml, performance / 11 min read
- SvelteKit 3 is here: Removes deprecated APIs, adopts native subpath imports and Vite 8 with Rolldown, and ships a migration command by Svelte / svelte, javascript, frontend / 2 min read
- Preact 11: After more than six years of work, the release brings a new hydration system, automatic ref forwarding and improved React compatibility by Preact / preact, javascript, frontend / 3 min read
- Effect 4.0: Rebuilt from the ground up, the library is now 5x smaller, 6.4x faster and uses 86% less memory by Effect / typescript, effect / 5 min read
🧰 Tools
- Atlas: Source control for coding agents that links commits to the sessions, prompts, tool calls and reasoning that produced them by pacifio / ai, agents, git
- Mobile MCP: One MCP interface for agents to automate native iOS and Android apps across simulators, emulators and real devices by mobile-next / mcp, mobile, testing
- Safe Not Safe: Checks PostgreSQL migrations locally in the browser with a WASM parser and deterministic rules, no upload or account by Safe Not Safe / postgres, databases, tools
- Floci: Free local emulators for AWS, Azure, GCP and OCI with real service engines and no cloud account needed by Floci / cloud, testing, tools
- OpenRig: Multi-agent harness that runs Claude Code and Codex sessions as one persistent team with YAML topologies, shared queues, a TUI and tmux recovery by mvschwarz / ai, agents, cli
- OpenAPPA: Open-source deterministic guardrail for coding agents that checks before every tool call whether data may flow to a destination, with security labels, tool contracts and remedy plans by Archestra / ai, agents, security
- jevgrep: CLI that searches code using natural language by David Zhang / cli, search, ai
- Size Limit: Performance budget tool that fails CI when a change pushes bundle size or load time over the limit by Andrey Sitnik / javascript, performance, ci
- MSW 3.0: API mocking library goes ESM-only, adds a Vite plugin, GraphQL subscriptions and intercepts page navigations and form submissions by MSW / testing, javascript, api
- Kafgres 0.2: A Kafka-compatible broker running inside Postgres, now at 700 MB/s write throughput on a single box by Raynor Elgie / postgres, kafka, streaming
- pg_raw_parse: Rust bindings to the Postgres parser with 20 to 60x faster parsing than pg_query.rs by PgDog / postgres, rust
- tinyjs: Desktop apps on the system native webview instead of a bundled browser, with a runtime of about 6 MB by tinyjs / desktop, javascript, tools
- celld: Self-hosted, distributed Durable Objects with a SQLite database per object, backed by object storage and compatible with Cloudflare config by celld / cloudflare, sqlite, serverless
- Moli: Headless browser built for agents by Lexmount / browser, agents, tools
- OpenShell: Private, sandboxed runtime for autonomous agents by NVIDIA / agents, security, sandbox
🤪 Fun
- We didn’t start the scaling: A website that pushes AI-generated content to its limit by Domenic Denicola / ai, web / 35 min read
- We ported the original Doom to SQL: The 1993 FPS logic and renderer running in SQL with 89 CTEs, returning each frame as a bitmap by CedarDB / sql, games / 30 min read
📺 Videos
- Andres Freund on where Linux helps and hurts Postgres: A candid Kernel Recipes talk on io_uring, huge pages, futexes and atomic writes by Andres Freund / postgres, linux
- WebMCP - Stop scraping the web, start calling it: A WebMCP spec co-author shows how explicit tool contracts replace agents guessing their way through pages by Patrick Brosset / ai, agents, web
- The unreasonable effectiveness of BM25 for agentic search: Why plain lexical search often beats embeddings when agents do the retrieval by Jo Kristian Bergum / ai, search