cuppa

brewed 03:49 AM
← May 12

Wednesday

may13

2026

May 14 →

the brief

Agentic workflows took center stage: Anthropic pushed long‑running execution in Claude Code and a Fast mode for Opus 4.7, while OpenAI expanded Codex with Symphony and true computer use. Vercel rounded out the stack with a web adapter for agents and tighter egress controls, and Microsoft showed what 100+ coordinated agents can do for security. It’s a day of autonomy meeting governance and shipping paths to production.

the poursit · sip · 12 items

alerts

(01)
  • anthropics/claude-code· feedMay 12, 09:09 PM

    Claude Code 2.1.140 fixes /goal hang

    Release resolves /goal silent hangs when hooks were restricted, improves tool subagent type matching, and fixes symlinked settings hot‑reload regressions—update if you rely on /goal or config hooks.

    v2.1.140 — What's changed Improved Agent tool subagent_type matching to accept case- and separator-insensitive values (e.g. "Code Reviewer" resolves to code-reviewer) Updated agent color palette Fixed /goal silently hanging when disableAllHooks or allowManagedHooksOnly is set — now shows a clear message instead of an indicator that never resolves Fixed a regression in settings hot-reload where symlinked settings files caused misattributed change events and spurious ConfigChange hooks Fixed cl...

    signal 9hype 1release_notesdevtoolsagent_frameworksource ↗

pulse

(08)
  • @ClaudeDevsMay 13, 12:00 AM

    Claude Code adds /goal for long runs

    New /goal lets you set completion conditions so Claude keeps working autonomously; paired with auto mode, /schedule, /loop, and stop hooks to reduce babysitting for multi‑step tasks.

    How do you keep Claude working until the job is done? Claude Code helps with this in a few ways, including one we shipped recently: /goal. pic.x.com/QtVPmwoKct

    signal 7hype 2claude_codefeature_releaseagent_workflowssource ↗
  • @unknownMay 12, 06:23 PM

    Opus 4.7 Fast mode lands

    Fast mode for Claude Opus 4.7 is in research preview on the API and in Claude Code, offering lower latency while retaining Opus‑class quality; API access via waitlist.

    Fast mode for Claude Opus 4.7 is now available in research preview on the API and in Claude Code.

    signal 7hype 1api_updateclaude_codefeature_launchsource ↗
  • @OpenAIDevsMay 12, 05:27 PM

    Codex Symphony runs task agents

    Symphony spins up a Codex agent for every open task, enabling persistent, parallel background work across your workspace rather than a single conversational thread.

    Symphony: every open task gets a running Codex agent x.com/2821275879/sta…

    signal 7hype 2agent_frameworkproduct_launchopenaisource ↗
  • @OpenAIDevsMay 12, 08:31 PM

    Codex gains true computer use

    OpenAI demos Codex that can click, type, and operate apps in the background across your Mac, moving from chat to hands‑off automation across multiple applications.

    Computer use lets Codex work across your apps without taking over your Mac. @AriX talks with @romainhuet about what changes when agents can click, type, and keep working in the background. pic.x.com/LdAmwtTXCY

  • @vercel_devMay 13, 01:08 AM

    Vercel Chat SDK web adapter

    Bring your server‑defined agent to the browser and stream responses with AISDK’s useChat hook, simplifying web delivery for agent apps.

    The web adapter for Chat SDK brings your agent to the browser. Define a bot on your server, then stream agent responses with @aisdk's 𝚞𝚜𝚎𝙲𝚑𝚊𝚝 hook. vercel.com/changelog/chat…

    signal 6hype 1sdk_updatebrowser_agentsreact_hooksource ↗
  • @vercel_devMay 12, 09:07 PM

    Vercel Sandbox firewall egress control

    Proxy and filter outbound agent requests, inject credentials granularly, and log/audit traffic—giving teams stricter least‑privilege and observability over agent network access.

    Vercel Sandbox firewall can now proxy and filter outbound requests. ▪︎ Control all agent traffic from your own proxy ▪︎ More granular credential injection ▪︎ Log and audit outgoing traffic vercel.com/changelog/verc…

    signal 6hype 2vercelproduct_updatesandbox_firewallsource ↗
  • @satyanadellaMay 13, 12:01 AM

    Microsoft unveils multi‑agent security

    Nadella highlights a multi‑model system of 100+ specialized agents that topped the CyberGym benchmark and helped find and fix 16 issues ahead of Patch Tuesday.

    Our new multi-model agentic security system brings together more than 100 specialized agents across frontier and custom models to find exploitable bugs, delivering top performance on the CyberGym benchmark. We used it ahead of Patch Tuesday to help find and fix 16

    signal 7hype 3agent_frameworkssecuritymicrosoftsource ↗
  • vercel/news· feedMay 12, 07:00 PM

    Vercel WAF adds NL rules

    Author custom firewall rules by describing intent—generate rate limits, blocks, redirects, and more without regex, then review and deploy from the dashboard.

    Create Vercel Firewall rules with natural language — now lets you create WAF custom rules using natural language. Describe the behavior you need and the dashboard will generate the rule.Vercel Firewall Visit the to try creating a rate-limiting rule:firewall custom rules page let you control traffic to your site by logging, blocking, challenging, rate limiting, or redirecting requests based on conditions like IP address, path, country, user agent, and more.WAF custom rules For example, you can...

findings

(01)
  • hn/frontpage· feedMay 12, 11:46 PM

    Linux idle tweak broke QUIC flows

    Cloudflare unpacks a “QUIC death spiral” rooted in a Linux power‑management optimization misclassifying idle, degrading performance—useful kernel‑protocol interaction lessons for systems engineers.

    When "idle" isn't idle: how a Linux kernel optimization became a QUIC bug — Article URL: https://blog.cloudflare.com/quic-death-spiral-fix/ Comments URL: https://news.ycombinator.com/item?id=48116064 Points: 20 # Comments: 1

    signal 7hype 1postmortemnetworkinglinux_kernelsource ↗

voices

(02)
  • @ctatedevMay 12, 03:04 PM

    Locking down PRs amid supply‑chain hits

    A maintainer weighs disabling external PR creation in favor of automation and provenance to blunt the surge of dependency and contributor supply‑chain incidents.

    As a follow-up to this I'm considering disabling PR creation from outside contributors on repos I maintain until we have better answers for the constant stream of supply chain and security incidents hitting the ecosystem Instead, I want to invest in automation that properly x.com/ctatedev/statu…

    signal 4hype 1open_sourcesecuritysoftware_supply_chainsource ↗
  • @nicbstmeMay 13, 01:02 AM

    /goal as primitive for long tasks

    Argues long‑running work needs harness‑level state—/goal acts as a simple initializer agent persisting across crashes, limited context, and days, beyond what base models can do alone.

    I'm glad to see /goal becoming the new primitive for long-running tasks. The model does not naturally persist across turns, context windows, sandboxes, process crashes, or days of work so it needs the help of the harness. I also love how simple it is. An "initializer agent" x.com/ClaudeDevs/sta…

    signal 5hype 1agent_frameworklong_running_taskspersistencesource ↗