THE AGENT ACTION FIREWALL TEST
Can your agent take a dangerous action without a receipt?
Paste an MCP server manifest, an OpenAPI spec, or a tool list. The fire drill flags every dangerous operation — money movement, data destruction, production deploy, permission change, bulk export, regulated override — that can run without an accountable human receipt.
Runs the same scanner as npx @emilia-protocol/fire-drill. Static assessment — verify the fix at runtime with EG-1 conformance.
If your agent can take an irreversible action without a receipt, you do not have control. You have hope.
EARNED EG-1? EMBED THE BADGE
