0.7.0 is a new audit pipeline - more agents and better tools, running in parallel. The headline is quality.
59.8% to 88.9% on EVMBench
The public EVMBench post measured 59.8% recall. The later attributed full run is 88.9%. Precision was not scored on that run. A full audit is still a multi-hour job.
Starting an audit is simpler
We simplified starting an audit. You answer a few questions - what to audit, which version to pin, scope, optional docs - then the audit runs.
Fewer duplicate findings
Similar findings collapse into groups, so the list is shorter and easier to read.
A reworked PDF report
We reworked the downloadable PDF report so the write-up you export matches the finding you already read in the product.
CLI and a skill for your agent
There is a Guardix CLI, and a skill that connects your coding agent to the audit it is sitting in - read findings, start a run, record a verdict.
Start a Code Audit from New audit. In the app, reopen this note anytime from the help menu.