What the review includes.
- Map up to three agreed untrusted-content ingestion paths and their connected action sinks.
- Run non-destructive, documented scenarios against the agreed test environment or supplied sanitized artifacts.
- Record observed behavior, exact reproduction steps, controls already present, and material limitations.
- Deliver one prioritized findings report, one control-plan readout, and one written clarification round.
View a clearly labelled synthetic sample deliverable before requesting the review.
Good fit — and not a fit.
Good fit
- Your agent reads email, tickets, documents, web pages, RAG content, or tool results.
- It can send, refund, cancel, update, write, or call another consequential tool.
- You can supply an authorized test workflow, sanitized artifacts, and a technical owner.
Not a fit
- You need certification, legal advice, exploit development, or a complete production penetration test.
- You cannot authorize the environment or safely isolate test actions.
- You need a guaranteed finding, guaranteed fix, or a claim that the system is “safe.”
What happens next.
- Fit check. You send the product URL and one input-to-action path. No secrets or customer data.
- Written scope. We confirm paths, access, owner, exclusions, start date, delivery date, fee, and payment schedule.
- Authorized review. Testing starts only after access and inputs are sufficient.
- Evidence handoff. You receive the report, prioritized controls, readout, and one clarification round.