Transaction-Integrity Check

The confused deputy that acts on state that already changed.

An agent READS a precondition fact — an account balance, an in-stock quantity, an approval grant, a permission, a price, a lock — then EXECUTES a side-effecting action that depends on that fact after the world changed between the check and the use. The approved amount was already disbursed, the item already sold, the approval already revoked, the balance already drained. Supply-chain grades the tool DEFINITION drifting, approval discipline grades whether an approval was obtained, business-logic grades static rules — none grade runtime world-STATE staleness. AgentSmack scores a TOCTOU state-drift transcript for exactly this check→act race — surfacing the read→invalidate→action lineage as evidence, mapped to OWASP LLM / NIST AI RMF coverage with paste-able remediation, a synthetic-attacker gauntlet, and audience report variants. Ground-truth re-validation before the action clears the finding. Load a sample to watch the gauge swing with no live infra.

Drive a live agent endpoint

Don't paste a transcript — give your agent a benign objective that depends on a precondition (an approval grant, a balance, an inventory count), point AgentSmack at your real staging endpoint, and watch it wire the already-disbursed funds again on a stale read after the world changed between turns. Each turn is a real HTTP round-trip; the precondition value is redacted by construction.

This POSTs to a real endpoint you control. localhost / private / link-local targets are refused server-side by the SSRF guard (re-validated before every request).

Auth headers (optional)