Key topics
- Webhook-triggered runs with structured payload schemas.
- Event-driven orchestration and queue-based processing.
- Chat integrations with scoped permissions and safe commands.
- SSO and service-to-service authentication patterns.
- Audit and observability integration (centralized logs and traces).
Common pitfalls
- Unstructured inputs that force the model to guess parameters.
- Exposing too many tools and endpoints to the agent runtime.
- Lack of correlation IDs and trace propagation across systems.
- Authentication shortcuts that bypass enterprise policies.
Recommended practices
- Standardize contracts (OpenAPI, JSON schemas) for tools and events.
- Apply least privilege and approval gates for write actions.
- Propagate correlation IDs through all calls and logs.
- Test integrations with failure injection and replay scenarios.
This page is intended to be actionable for engineering teams. For platform-specific details, cross-reference /platform/agents, /platform/orchestration, and /platform/knowledge.