The Static Agent Web

Tools for AI agents are files, not servers.

Publishstatic files
→
Discoverthe registry
→
Verifyby hash
→
Executewasm sandbox
→
Knowverified RAG
→
Learnshared memory

No server on any link. Every step is a static file a browser or a runtime can verify.

Start here — pick your path

The trust model, in four rings

Integrity

SHA-256: the audited bytes are the executed bytes.

Authenticity

ed25519 signatures + key pinning.

Freshness

Signed "still true", voided on edit, expiring.

Stability

A lockfile: the publisher can't change a tool silently.

These four rings verify the skill itself. Running it safely is the runtime's job too — see mcpwasm's sandboxing.

The pieces