the brief
Today was about running and governing agents at scale. Cloudflare unveiled an end‑to‑end agent operations stack—console, lifecycle, CI/CD, local tracing, and wallets—while LLM 0.32 and Hugging Face’s LFM2.5 push local and auditable agents forward. Security stayed front‑and‑center with npm token rotations and a massive supply‑chain incident, plus Mistral’s safety classifier and OpenAI’s cyber evaluation postmortem.
Following a contained incident, npm will rotate write‑scoped Granular Access Tokens that bypass 2FA; maintainers should upgrade npm CLI v12+ and adopt Trusted Publishing.
npm is rotating write-scoped npm Granular Access Tokens that bypass 2FA as a precaution following a now-contained security incident. This doesn't affect GitHub personal access tokens. Maintainers should upgrade the npm CLI to v12+ and consider Trusted Publishing. docs.npmjs.com/trusted-publ...
Researchers report a self‑propagating malware campaign compromising popular packages like Keyv, Cacheable, and flat‑cache (2B monthly downloads), underscoring urgent supply‑chain hygiene.
Researchers: ChainDrop, a Shai-Hulud-based worm, has compromised 1,300+ npm packages, like Keyv, Cacheable, and flat-cache, with a combined 2B monthly downloads (Bill Toulas/BleepingComputer) — Bill Toulas / BleepingComputer: Researchers: ChainDrop, a Shai-Hulud-based worm, has compromised 1,300+ npm packages, like Keyv, Cacheable, and flat-cache, with a combined 2B monthly downloads — Self-propagating malware named ‘ChainDrop’ has compromised more than 1,300 packages with a combined 2 billio...
Low‑code LLM orchestration tool Flowise is sunsetting, prompting users to migrate workflows and data to alternatives.
Flowise Is Shutting Down — Article URL: https://flowiseai.com/sunset Comments URL: https://news.ycombinator.com/item?id=49176920 Points: 8 # Comments: 3
Simon Willison’s CLI gains visible reasoning traces, server-side tools, OpenAI Responses support, and content‑addressable SQLite logs, making local model ops and auditing more robust for agent workflows.
New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging — <p>I released <a href="https://llm.datasette.io/en/stable/changelog.html#v0-32">LLM 0.32</a> this morning, the most significant new version of LLM since the initial launch of the project. The new version includes support for visible reasoning traces, server-side provider tools, redesigned content-addressable SQLite logs, new models, and new features enabled by the OpenAI Responses ...
Plugin updates unlock new Claude Fable, Sonnet, and Opus 5 variants plus features enabled by LLM 0.32, tightening Anthropic integration for power users and agents.
llm-anthropic 0.26 — <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-anthropic/releases/tag/0.26">llm-anthropic 0.26</a></p> <p>Includes new features enabled by <a href="https://simonwillison.net/2026/Aug/4/new-release-of-llm/">LLM 0.32</a>:</p> <blockquote> <ul> <li>New models: <code>claude-fable-5</code>, <code>claude-sonnet-5</code>, and <code>claude-opus-5</code>. <a href="https://github.com/simonw/llm-anthropic/issues/75">#75</a>, <a href="https://github.com/simonw/ll...
Lightweight 2.6B agentic controller aims to run tool‑using assistants locally across devices, bringing lower‑latency, private workflows without cloud dependence.
Deploy local agents everywhere with LFM2.5-2.6B
New dashboard aggregates and monitors deployed agent sessions, surfacing health and performance to help teams operate AI agents at scale across Workers.
Introducing: Cloudflare Agents — Cloudflare Agents brings all of your deployed agent sessions into a single experience, surfacing key information and insights into how your agents perform at scale.
Cloudflare proposes an Agent Development Lifecycle with underlying primitives so teams can build, review, deploy, and maintain code‑writing agents safely and repeatably.
The Agent Development Lifecycle has arrived on Cloudflare — Agents can write code faster than teams can review, deploy, and maintain it. Today we’re introducing the Agent Development Lifecycle and the Cloudflare primitives that underpin it.
Workflows, Artifacts, and a CI SDK let you run customizable, sandboxed pipelines in Workers with TypeScript steps and self‑healing agent assistance, reducing YAML sprawl.
Run CI/CD for millions of repos — on your platform, on Cloudflare — Learn how to build customizable, sandboxed CI/CD pipelines natively on Cloudflare using Workflows, Artifacts, and the CI SDK. We walk through replacing complex YAML configurations with TypeScript workflow steps and self-healing AI agents.
wrangler dev emits structured traces per request, giving coding agents a one‑shot API to pinpoint failures locally before deploys.
Your agent can now debug Workers with local tracing — wrangler dev now produces structured traces for every local request. Your coding agent can hit a single API to pinpoint exactly what failed and why — no deployment required.
Programmable wallets with x402 enable agents to pay for APIs and content with verifiable identity and guardrails, moving agentic commerce onto the web.
Announcing Cloudflare Wallets: The programmable wallet for the agentic Internet — Cloudflare Wallets will provide AI agents with native payments and verifiable identity on the web. Using the x402 protocol, agents can autonomously purchase APIs and content within clear safety guardrails.
A Python package brings MiniMax‑H3’s omni‑modal text‑image‑audio‑video capability to Apple’s MLX, enabling 15‑second video generation with audio on Macs.
PipeNetwork/minimax-h3-mlx — <p><strong><a href="https://github.com/PipeNetwork/minimax-h3-mlx">PipeNetwork/minimax-h3-mlx</a></strong></p> MiniMax released <a href="https://huggingface.co/MiniMaxAI/MiniMax-H3">MiniMax-H3</a> two days ago - they describe it as a "a general-purpose, omni-modal generative system", which in practice means it accepts text, images, audio and video and can use them to generate up to 15 second video clips with audio included.</p> <p>This Python package ports it to M...
Shieldstral is a 3B multimodal safety classifier, Apache‑2.0 licensed, claiming parity with models up to 7× larger for text safety tasks.
Mistral releases Shieldstral, a 3B multimodal safety classifier that it says matches models up to 7x its size on text safety, available under Apache 2.0 (Mistral AI Blog) — Mistral AI Blog: Mistral releases Shieldstral, a 3B multimodal safety classifier that it says matches models up to 7x its size on text safety, available under Apache 2.0 — Every product that ships a model needs to answer questions like these — but the right answer depends on the product, the audience, and the moment.
OpenAI discloses two incidents from independent evaluations, how they were contained, and new safeguards for future third‑party testing of advanced models.
Third-party cyber evaluations involving OpenAI models — OpenAI explains recent third-party cybersecurity evaluation incidents and outlines new safeguards to strengthen AI model testing and evaluation.
Release fixes allowlists and isolation to prevent destructive git commands from background subagents and tighten tool restrictions across sessions.
v2.1.222 — What's changed Fixed worktree-isolated sessions and their subagents being able to run destructive git commands against the main checkout; isolation now applies to file edits and Bash in every session type Fixed PreToolUse auto-allow hooks bypassing tool restrictions in background agent tasks (summaries, compaction, renames) Fixed /usage-credits on Team and Enterprise showing "you've already sent a usage credit request" for members whose earlier request was dismissed, blocking them ...
Cocktail‑Talker introduces Turn Action GRPO to model multi‑speaker dialogue under background noise, exploring strategies for robust conversational agents in social settings.
Cocktail-Talker: Multi-Speaker Dialog Modeling in Noisy Social Environments with Turn Action GRPO arxiv.org/abs/2607.27756
Latent Space reconstructs memory, proactivity, scheduling, browsing, plugins, skills, and tools in ChatGPT Work, offering a clear mental model for enterprise agent behavior.
Unpacking ChatGPT Work: the Agent for a Billion Users — An external reconstruction of how Memory, Proactivity, Scheduling, Browser Use, Plugins, Skills and Tools work in the new ChatGPT Work.
Theo Mirr argues Codex and Claude act like terminals; resource leaks belong to spawned processes, not the agent itself—avoid misleading screenshots and blame.
You guys do realize that Codex and Claude are basically acting as terminals, right? If your dev server has a memory leak, and Codex spawns it, the memory leak will be credited to Codex. Imagine posting a screenshot of Ghostty using 100gb of RAM and saying it has a memory leak
Cloudflare describes Codex, a governed body of engineering standards consumed by agents to enforce consistency across code, specs, and incident reports with structured RFCs and reviews.
How Cloudflare enforces engineering standards using AI — We created the Cloudflare Codex, a governed body of engineering standards that AI agents consume across the development lifecycle. By pairing structured RFCs with agentic reviews, teams automatically enforce consistency across code, specs, and incident reports.
