Code review with the evidence in view
Ask the model to explain a function, identify security-relevant assumptions or organise a review checklist. Provide only code you are authorised to process. Require file and line references; verify each suggested issue before it becomes a finding.
Use Core for routine analysis and evaluate Large on more involved reasoning tasks. Compare models.
Triage authorised assessment output
Turn a supplied set of scan results or observations into an ordered review queue. Ask the model to separate confirmed evidence from hypotheses and list the next safe validation steps within the agreed scope. Automated output is a lead, not proof of a vulnerability.
Draft clear client-facing explanations
Start from findings you have already validated. Draft the impact, affected component, remediation and retest criteria in language the client can use. Keep technical evidence distinct from the executive summary, and verify every factual claim.
Track usage per client
Use a separate engagement key for each assignment, give it a budget and expiry, then export a date-bounded usage statement. An optional markup column helps prepare your own invoice without changing the API charge.
Read the engagement workflow or try the authorised-work templates.