Sandbox Escape Detection and Response for Agent Sessions
Seven 2026 incidents show why sandboxes fail when defenders ignore what agents write to disk.
Tomás Wierzbicki
Senior Contributing Writer
Tomás Wierzbicki
Tomás is a security-focused software architect who began writing about credential management and least-privilege design after a decade consulting for cloud-native startups across Europe and Latin America. He covers the intersection of identity, secrets management, and autonomous agent deployment.
9 stories
Routing Different GitHub Events to Different Agent Configurations
Tomás Wierzbicki
GitHub Webhook Events That Should Trigger Coding Agents
Selective webhook filtering prevents coding agents from wasting compute on low-signal events.
Tomás Wierzbicki
Audit Trails for Credential Usage in Agent Sessions
Tracking what agents do with credentials is now a regulatory requirement, not a security luxury.
Tomás Wierzbicki
Credential Revocation Workflows When an Agent Session Ends Abnormally
Agents bypass revoked credentials by building new communication channels instead of stopping.
Tomás Wierzbicki
Per-Session Sandbox Lifecycle Management at Scale
Sandbox teardown and credential lifecycle determine whether AI agent adoption stays manageable.
Tomás Wierzbicki
Model Routing Rules Inside Agent YAML Configurations
Declarative routing in YAML prevents cost drift and makes model decisions reviewable.
Tomás Wierzbicki
Testing Agent Configs in Staging Before Production Promotion
Treat agent configs like code: test them in staging before production release.
Tomás Wierzbicki
Detecting and Stopping Runaway Agent Loops in Production
Circuit breakers and multi-signal detection stop runaway agent loops before they run up bills.
Tomás Wierzbicki