Agent Plugins 1.0: The Write-Once Era Begins

#agenticai #aiagents #aws #opensource

AWS, Cursor, Microsoft, OpenAI, and Vercel agreed on something this week. That alone is news.

Agent Plugins 1.0 is now published. Google joined as Core Maintainer on day one. The spec packages Agent Skills and MCP servers into a single portable directory with a plugin.json at the root. Build once. Deploy to ChatGPT, Codex, Cursor, GitHub Copilot, Kiro, and VS Code.

That list covers most of where real developers spend their time. Which is the point.

What's in Issue #25

Agent Plugins 1.0 — the write-once, run-anywhere spec for agent skills and MCP servers
Three new harnesses in one week — Amazon's Kiro Crew (39K internal users, MIT), YC's QM (11K stars), Meta's Muse Code (persistent subagents that survive crashes)
Docker Sandboxes — one microVM per agent, zero host risk. #1 on HN at launch
Prime Agent — one tool (IPython kernel), 95.5% on ARC-AGI-3. Above human expert baseline
AWS Dogwood (Tool of the Week) — temporal policies for agent tool calls. Loop detection at the infrastructure layer. Apache 2.0
DynamoDB Vector Search GA — semantic retrieval over agent memory, no separate vector DB
Paper Breakdown — most visual tool calls don't change the answer. 6 models, 5 benchmarks, two failure modes
Hot Take — Claude Code auto mode is the right call and everyone's wrong about why