World2Agent wants to standardize your agent's senses

What it does

World2Agent (W2A) is an open protocol that defines a common schema for sensors feeding real-time data into AI agents. Sensors—ordinary npm packages—watch external sources like Hacker News, frontier AI labs, or weather APIs, then emit structured signals that any compliant runtime can consume. The project ships with plugins for Claude Code, Hermes, and OpenClaw, effectively giving those agents a standardized nervous system without custom plumbing for every feed.

The interesting bit

W2A treats perception as infrastructure rather than a one-off integration. Because every sensor speaks the same schema, swapping “watch the markets” for “watch the weather” is closer to changing a USB peripheral than rewriting a scraper—your agent only needs to know how to listen, not where the data originates.

Key highlights

• Shared signal schema makes sensors interchangeable across agent runtimes
• Native plugins for Claude Code, Hermes, and OpenClaw
• SensorHub catalogs official and community sensors by category (markets, news, production alerts, etc.)
• New sensors can be scaffolded in roughly 50 lines and published to npm
• Roadmap includes a graph layer for composing multi-sensor signals before they reach the agent

Caveats

• The README carries an explicit security warning: a malicious sensor is effectively an untrusted instruction source, so audit before installing
• The graph-layer signal composition feature is currently only an RFC, not implemented
• The Claude Code integration currently requires loading via a --dangerously-load-development-channels flag, signaling an experimental path

Verdict

Give it a spin if you’re running proactive agents that need to react to live external feeds and you’re tired of maintaining one-off scrapers. Pass if your agents are fully self-contained or you aren’t comfortable vetting third-party npm packages that can directly drive agent behavior.