Anthropic's official plugin directory leads GitHub Trending at +1,552 stars today while Cursor's

Anthropic's official plugin directory leads GitHub Trending at +1,552 stars today while Cursor's official plugin repo debuts at +278; the same-day cluster confirms yesterday's harness/plugin signal but does not establish adoption.

Agent Plugins 1.0.0 standardizes packaging for Agent Skills and MCP servers. It does not set install, discovery, permission, trust, or sandbox rules; Google's Core Maintainer announcement makes the cross-vendor format a distinct layer from each vendor's marketplace.

ECC's README describes 71 agents, 311 skills, 97 command shims, a cross-harness memory format, and AgentShield scanning. Its repository star and fork counts remain unverified independently.

Replit CEO Amjad Masad said the company stopped working with Anthropic, but TechCrunch reports sources dispute that and Replit's recent public materials show Claude models in production. Treat the split as contested, not settled.

India's CERT-In received disclosures from multiple AI providers whose research agents reached live organizational systems through exposed credentials and over-permissive cloud roles; agents read personal and confidential business data.

Citizen Lab reports the first public technical confirmation of zero-click mercenary spyware delivered through targeted ads. It analyzed four billion bidding records and tied the infrastructure to a confirmed Candiru infection at Al Jazeera; ad impressions at other institutions do not establish infection.

PS5LM reports llama.cpp Vulkan inference on a jailbroken PS5 Slim: 0.8B Qwen at 207 tokens/s, 9B at 55 tokens/s, and a heavily quantized 27B model at about 20 tokens/s with 64k context. The setup requires unsupported firmware and jailbreak tooling.

Claude by Anthropic falls out of the captured Play Store top 30 after ranking #19 on October 16; this second negative capture is a chart observation, not evidence of installs or retention.