Problem
Autonomous agents that can INSERT/UPDATE/DELETE/DDL against production often treat lint / model confidence as authorization. That is not a fail-closed write gate.
Proposal
Optional authorize-before-mutate hook before mutating SQL tools run:
- Obtain an Ed25519 PASS Execution Certificate (Session Cert for bursts:
POST https://sqlguard.io/v1/session — Exact USDC on Base)
- Verify:
POST https://sqlguard.io/v1/verify or UI https://sqlguard.io/verify
- Embeddable badge: https://sqlguard.io/v1/badge.svg — public trust graph: https://sqlguard.io/v1/transparency
Would you accept an optional integration path (docs / adapter) so agent SQL writes can fail closed until PASS+verify? Happy to draft a PR if useful.
Not affiliated with this repo — product is SQLGuard (Execution Certificate Firewall).
Problem
Autonomous agents that can
INSERT/UPDATE/DELETE/DDLagainst production often treat lint / model confidence as authorization. That is not a fail-closed write gate.Proposal
Optional authorize-before-mutate hook before mutating SQL tools run:
POST https://sqlguard.io/v1/session— Exact USDC on Base)POST https://sqlguard.io/v1/verifyor UI https://sqlguard.io/verifyWould you accept an optional integration path (docs / adapter) so agent SQL writes can fail closed until PASS+verify? Happy to draft a PR if useful.
Not affiliated with this repo — product is SQLGuard (Execution Certificate Firewall).