Chained attacks, graded like every other test
A preset of the scanner built from its adversarial modules. Its chained tests build a conversation from authored prefixes before the final attack, then grade the answer with the same published rules.
In this release
You can use this in your workspace today.
- Adversarial presetAn adversarial module preset whose chained tests replay authored prefixes, without an adaptive multi-turn engine.Live
Chained tests
Each chained test sends its authored opening turns and the target's real replies before the final attack, instead of a single prompt.
No adaptive engine
The chains are fixed. Secuvon does not run an adaptive attacker that rewrites its next turn from your agent's answers.
Graded evidence
Every chained test is graded by its published rule and reported with the full transcript, like any other finding.
Unsupported, not faked
A target that cannot hold a conversation reports chained tests as unsupported instead of pretending they ran.
Open
Send the authored opening turns and keep the target's real replies.
Attack
Send the final attack in that conversation.
Grade
Grade the final answer by the test's published rule, with the transcript as evidence.