AI triage for security scanner findings

Crongent triages security scanner output with an AI agent that reads the code behind each finding at the exact commit and returns a severity, a verdict (real, needs review, false positive) and a fix hint, with every file it read in the run log.

The problem

Scanners produce hundreds of findings and most are noise. Engineers stop reading them, so the real one ships.

How to set it up in Crongent

  1. Run your scanners in CI as usual.
  2. Start a Crongent triage run through the API with the commit and scanner output.
  3. The agent reads the code behind each finding through GitHub and returns its verdicts.
  4. Your release gate reads the result from the API or webhook.

Example routine

Name
Security triage
Trigger
API trigger on release request
Model
anthropic/claude-sonnet-5.5
Instructions
For each scanner finding and each changed file, decide if it is exploitable. Cite file and line. Never report a finding you did not read.

Who uses it

Frequently asked questions

Is AI triage a replacement for a security review?
No. It is an advisory first pass that cuts noise so humans spend time on the findings that matter.

Related use cases