Anthropic has donated Agent Skills to the Agentic AI Foundation under the Linux Foundation. For anyone deciding whether skills are a Claude feature or a format worth standardising on, that settles it.
It's a quiet day. The rest is agent plumbing: credentials, runtimes and KV cache. No repos or videos cleared the bar, so those sections are empty.
Model & provider releases
Anthropic donates Agent Skills to the Agentic AI Foundation — Simon Willison / Anthropic
The standard now sits with the AAIF under the Linux Foundation, which Anthropic co-founded with Block and OpenAI. That makes skills a portable format, not vendor lock-in. If you've been holding off on writing an internal skill library, stop waiting. https://simonwillison.net/2026/Oct/22/donating-agent-skills/
1Password lets AI agents log in without seeing your password — TechCrunch
Credentials that never enter the context window can't be leaked by a prompt-injected agent. Before you wire it into anything customer-facing, I'd check how the resulting sessions are scoped and revoked. https://techcrunch.com/2026/10/22/1password-launches-a-way-for-ai-agents-to-log-in-on-your-behalf-without-seeing-your-password/
Actors for Agents: Patterns for Production AI — InfoQ
Agents are distributed systems; this talk treats them that way. https://www.infoq.com/presentations/actors-for-production-ai-agents/
Research worth reading
Cache Engineering for Agents
On long agent trajectories, the authors argue KV cache I/O and storage become the dominant cost, and trajectories that outgrow the context window need controlled truncation. If you run long-horizon agents, keep prefixes cache-stable and treat context compaction as a deliberate design choice. Model choice matters less here than most teams assume. https://huggingface.co/papers/2610.28577
Pick one agent you already run and look at it through the lens of these three. Can it touch a credential directly? Does its state survive a crash? Is its prompt prefix stable enough to hit the cache? Whichever answer is worst is your next sprint.
From the field
A practitioner note from the build side.
Most Multi-Agent Systems Should Be One Agent and a State Machine
Most multi-agent architectures we're asked to build should be one scoped agent plus deterministic orchestration. When handoffs earn their keep, and a test to decide.
— Daniel · usqrd.com · reply to this email, I read everything