Security Engineer

1. 1 Parse incident signals Claude Opus 4.8 620ms

   Egress is bursty, not steady — inconsistent with scheduled backups or CDN sync. Treating it as unplanned traffic.

2. 2 Correlate with threat intelligence GPT-5 1140ms

   2 of the 3 destinations match an exfiltration cluster reported 6 days ago. The third is unclassified — not assuming it is benign.

3. 3 Assess blast radius Claude Opus 4.8 880ms

   svc-api-deploy can read 4 databases. Customer PII is in scope, so this is a disclosure risk, not just an availability one.

4. 4 Draft remediation plan GPT-5 1320ms

   Revoke credentials before isolating the host — cuts the attacker’s access without taking the API offline for customers.

5. 5 Check against policy guardrails Gemini 3 Pro 540ms

   Plan stays inside read-only permissions. Severity is High, so the on-call lead must be paged rather than auto-remediated.

Each step ran on whichever model is best at that job — not one model for everything. The worker is the workflow.

• Read: incident logs, asset inventory and threat-intel feeds — read-only, and never the credential stores
• Write: none; it drafts a triage decision and remediation plan for a human to approve and run
• Escalate: the on-call security lead the moment severity reaches High, without waiting for the full analysis
• Never: touch production credentials, secrets or infrastructure directly
• Never: take a remediating action itself — every change is proposed, never applied

• Classified as a likely exfiltration — 2 of 3 IPs match a known cluster.
• Proposed: revoke svc-api-deploy credentials, then isolate the host. Paged the on-call lead.
• Took no action itself. Zero production credentials touched.

private rules were taught in one workspace and never leave it. general rules are true about the work, so every copy of this worker gets them.