Agents
Custom agents
Extend the platform with domain-specific testing intelligence.
Overview
Custom agents extend the Zof catalog with domain-specific validation using the agent SDK. They run under organizational policy with declared capabilities, runtime requirements, and labels for Agent Console targeting.
Custom agents receive jobs from the same orchestration plane as catalog agents; results appear in Runs and Test Health with identical audit trails.
Who should read this
- Platform teams and specialized QA engineers building proprietary validation for regulated or unique application stacks.
Prerequisites
- Agent SDK access approved by administrator
- Runtime environment matching declared capabilities (cloud or endpoint)
- Security review for code executing in customer infrastructure
When to use this workflow
- Onboarding new team members to Zof terminology and workflows
- Authoring internal runbooks aligned with Console labels
- Designing CI/CD or webhook integrations against documented behavior
Step-by-step procedure
Define manifest
Declare agent name, category, capabilities, supported environments, and resource limits.
Submit manifest for administrator catalog approval.
Implement validation logic
Implement job handler receiving test payload and returning structured pass/fail results.
Report progress heartbeats for long-running jobs.
Deploy runtime
Package agent for cloud registry or endpoint agent installation.
Verify connectivity from Agent Console fleet view.
Enable for organization
Admin approves catalog entry in Automation → AI Agents.
Operators target agent via labels in run configuration.
Key concepts
- Agent manifest
- YAML or JSON declaring identity, capabilities, and policy constraints.
- Job capsule
- Signed execution package delivered to endpoint agents.
Best practices
- Version agent images and manifest together
- Human review custom agent output before release-critical gates
- Log agent version in run metadata for reproducibility
Was this page helpful?
