Jailbreak Regression Suite

Turn every jailbreak you find into a permanent regression test, so safety boundaries never silently erode across model and prompt changes.

0 runs0 forks0

Chain

Inputs

Step 1

flow-jailbreak-regression-suite-1-assemble-the-jailbreak-corpus

Step 2

flow-jailbreak-regression-suite-2-build-reliable-compliance-evaluation

Step 3

flow-jailbreak-regression-suite-3-version-gate-and-grow-the-suite

Final output

No inputs required — hit Run to stream each step in order.

Each step streams from its model and feeds the next. BYOK — you pay your provider directly. Add a key in Settings if a step needs one.

Step 1 · Collect the attacks that matter for your agent's boundaries.

flow-jailbreak-regression-suite-1-assemble-the-jailbreak-corpus

Claude Sonnet 4.6→ {{step_1}}

Awaiting upstream steps…

Step 2 · The hard part is telling a real refusal from a soft yes.

flow-jailbreak-regression-suite-2-build-reliable-compliance-evaluation

Claude Sonnet 4.6→ {{step_2}}

Awaiting upstream steps…

Step 3 · A safety net that is not tied to CI and not growing will fail you quietly.

flow-jailbreak-regression-suite-3-version-gate-and-grow-the-suite

Claude Sonnet 4.6→ {{step_3}}

Awaiting upstream steps…

Turn every jailbreak you find into a permanent regression test, so safety boundaries never silently erode across model and prompt changes.

1
Collect the attacks that matter for your agent's boundaries.
Runs flow-jailbreak-regression-suite-1-assemble-the-jailbreak-corpus on Claude Sonnet 4.6
2
The hard part is telling a real refusal from a soft yes.
Runs flow-jailbreak-regression-suite-2-build-reliable-compliance-evaluation on Claude Sonnet 4.6
3
A safety net that is not tied to CI and not growing will fail you quietly.
Runs flow-jailbreak-regression-suite-3-version-gate-and-grow-the-suite on Claude Sonnet 4.6

•Recurring tasks you currently copy-paste between tools.
•When a single prompt isn't enough — you need a chain where output(n) feeds prompt(n+1).
•Fork and adapt if your inputs differ materially from the author's use case.

Claude Sonnet 4.6