🔒New Plugins Reduce False Positives in Patch Development
Fewer False Positives in Patching
TL;DR
Trail of Bits introduces two plugins to streamline patch development and reduce false positives in regression checks. The plugins, patching-bugs and regression-hunt, aim to improve security and efficiency in maintaining software.
Trail of Bits has launched two new plugins, patching-bugs and regression-hunt, designed to streamline patch development and minimize false positives in regression checks. These tools help engineers develop fixes for reported bugs and ensure that changes don't break existing functionality. In preliminary tests, regression-hunt produced 25% fewer false positives compared to a baseline Codex agent. The plugins are part of a structured workflow that provides maintainers with evidence and independent checks, improving patching hygiene. Engineers can direct the work and review results, while maintainers can supply security assumptions and patching requirements.
Key Points
The patching-bugs plugin reproduces reported bugs, develops fixes, and tests them, streamlining best practices from thousands of agent sessions.
The regression-hunt plugin checks for changes that break previously working behavior, confirming 21 regressions from over 130 PRs merged.
In preliminary evaluations, regression-hunt produced 25% fewer false positives compared to a baseline Codex agent.
The plugins incorporate a threat model if available, passing this document as part of the global context for each individual agent step.
The plugins will continue to evolve, with data collected on their usage in Patch the Planet work.
Why It Matters
If you're maintaining a critical software system, these plugins can significantly reduce the time spent on false positives and improve the reliability of your patching process. For instance, the regression-hunt plugin confirmed 21 regressions among 58 candidates in a single run, with a precision of ~36%. This means fewer wasted hours and more secure software.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,569 builders reading daily.