Alright, I know this forum is usually for templates, but I wanted to share my *process* and key security questions from our recent switch. We moved our marketing automation from a single, monolithic platform to a composable stack using Claw agents. The security review was the biggest hurdle, and our internal notes might help anyone building a similar rubric.
Our core concern was data flow and access control. With a monolith, it's one vendor. With agents handling data enrichment, lead scoring, and email triggers independently, every connection is a potential leak point. Here's what we drilled into:
**For Each Agent Connection:**
* **Authentication & Secrets:** Where are API keys stored? How are they rotated? (We mandated a dedicated secrets manager, not environment variables.)
* **Data in Transit:** Is encryption (TLS 1.3+) enforced for *all* calls, not just "supported"?
* **Data Minimization:** Does the agent request only the specific fields needed? We found one enrichment agent asking for full lead objects when it only needed company size and industry—big red flag.
* **Logging & Auditing:** What PII is logged by default? Can we suppress it? We required audit logs for all data access events.
**The Big Picture Items:**
* **Vendor Security Posture:** SOC 2 Type II was table stakes. We also asked for their incident response playbook.
* **Data Residency & Processing:** Mapping where data was processed at each stage was a nightmare, but crucial for GDPR.
* **Default Deny:** We configured our network so agents had *no* internet access by default, then whitelisted only the necessary endpoints.
The win was granular control. Instead of trusting one massive platform with everything, we can now isolate risks. If the scoring agent is compromised, it doesn't mean the email agent is too. But it requires way more diligence.
Would love to compare notes—anyone else gone through a similar decomposition? What was on your security checklist that I missed?
—Kate