Claude Mythos 5.2, Workflow Studio and your own AI
- QuillShield now supports Claude Mythos 5.2, and QuillAudits is one of the security teams with access to it.
- Choose how an audit runs. Four official workflows focus on entry-point attack paths, access-control consistency, oracle and price manipulation, or accounting invariants. The QuillShield Audit Agent stays the default, and every workflow costs the same as a full audit of the same code.
- Workflow Studio builds a custom workflow from a plain-language description, and an assistant refines it as you chat. Every change is a numbered draft you can undo, and a saved version never changes, so an audit always runs the exact version it started with.
- Workflows can run follow-ups on each finding, such as a Foundry proof of concept for High and Critical findings. A proof of concept counts as passed only when QuillShield re-runs it in a clean copy of your code.
- Publish a workflow to the Workflow Marketplace for other teams to run, copy and review. QuillShield reviews every published version before other teams can see it.
- Where your workspace offers it, a custom workflow can run on your own provider key (OpenRouter, Anthropic, OpenAI, xAI or DeepSeek) or on your team's Claude or ChatGPT plan. If that AI stops mid-audit, the audit pauses, and Resume audit continues it at no extra charge.
- Database connections now require TLS with certificate checks, and QuillShield blocks abusive traffic automatically.