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
- Run your scanners in CI as usual.
- Start a Crongent triage run through the API with the commit and scanner output.
- The agent reads the code behind each finding through GitHub and returns its verdicts.
- 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.