Custom Agent Lifecycle Management for AI Catalog

Everyone can contribute. Help move this issue forward while earning points, leveling up and collecting rewards.

• Close this issue

Summary

Define and implement a comprehensive lifecycle management system for custom agents in the GitLab AI Catalog. This roadmap item establishes the states, transitions, and governance controls that enable organizations to safely create, test, approve, and manage custom agents throughout their operational lifetime.

Problem Statement

Organizations need a structured approach to managing custom agents across their lifecycle to ensure:

• Safety & Governance: Agents are reviewed and approved before enterprise use
• Visibility: Clear understanding of agent status and maturity
• Control: Ability to manage agent versions, updates, and decommissioning
• Compliance: Audit trails and approval workflows for regulated environments

Agent Lifecycle States

1. Creation

• Agent is initially created by a developer or team
• Agent exists in draft/development state
• Not yet available for general use
• Creator can iterate and test locally

2. WIP (Work In Progress) / Testing

• Agent is being actively developed and tested
• Available for limited testing within a team or project
• Not recommended for production use
• Clear labeling to indicate experimental status
• Feedback loop for improvements

3. Approval

• Agent undergoes review process for enterprise use
• Security, compliance, and functionality validation
• Designated approvers (e.g., security team, platform team) review agent
• Approval criteria defined and documented
• Audit trail of approval decisions

4. Approved

• Agent has passed approval process
• Available for broader organizational use
• Listed in AI Catalog as production-ready
• Clear documentation and support expectations
• Version tracking and release notes

5. Update / Maintenance

• Agent receives updates, patches, or enhancements
• Updates follow same approval workflow as initial creation
• Version management and changelog tracking
• Backward compatibility considerations
• Communication of changes to users

6. Decommissioned

• Agent is retired or no longer supported
• Clear deprecation timeline communicated
• Migration path provided for dependent workflows
• Archived for historical/audit purposes
• Removal from active AI Catalog

Key Features Required

Governance & Approval

• Configurable approval workflows (who can approve, approval criteria)
• Role-based access control (creator, approver, admin)
• Approval request templates and checklists
• Audit trail of all lifecycle transitions
• Approval history and decision documentation

Visibility & Status Management

• Clear status indicators for each lifecycle state
• Status badges in AI Catalog
• Filtering/searching by agent status
• Dashboard showing agent lifecycle metrics
• Notifications for status changes

Version Management

• Semantic versioning support
• Version history and changelog tracking
• Ability to rollback to previous versions
• Breaking change indicators
• Deprecation warnings for older versions

Documentation & Communication

• Agent metadata (description, owner, support contact)
• Usage guidelines and best practices
• Known limitations and constraints
• Support SLA definitions
• Deprecation notices and migration guides

Permissions & Controls

• Creator permissions (edit, delete, submit for approval)
• Approver permissions (review, approve, reject)
• Admin permissions (override, force state changes)
• Organization-level policy enforcement
• Project/group-level agent restrictions

Related Issues

• #561358 - AI Catalog UX Acceptance Criteria (Creation and Editing section)
• #515283 - FY26 Duo Workflow investment themes and Road-Map (Q2-Q4 initiatives)
• #543161 - GitLab's Knowledge Moat & Interoperability Strategy

Success Metrics

• Agents can transition through all lifecycle states
• Approval workflow completion time < 48 hours
• 100% audit trail coverage for all state transitions
• User satisfaction with approval process > 4/5
• Adoption of custom agents increases by 50% post-launch
• Support ticket volume for agent issues decreases by 30%

Acceptance Criteria

• Lifecycle states are clearly defined and documented
• UI/UX for managing agent lifecycle is intuitive
• Approval workflow is configurable per organization
• All state transitions are logged and auditable
• Status is visible throughout AI Catalog and related interfaces
• Version management supports rollback and history
• Deprecation and decommissioning workflows are clear
• Documentation covers all lifecycle states and transitions

Implementation Considerations

Phase 1: Foundation

• Define lifecycle states and transitions
• Implement basic status management
• Create approval workflow framework

Phase 2: Governance

• Add role-based access controls
• Implement approval workflows
• Build audit logging

Phase 3: Enhancement

• Version management system
• Advanced filtering and search
• Deprecation and decommissioning workflows

Phase 4: Optimization

• Dashboard and analytics
• Automated notifications
• Integration with compliance tools

Notes

This issue consolidates lifecycle management requirements from:

• AI Catalog MVP acceptance criteria
• FY26 Duo Workflow roadmap
• Customer feedback on agent governance needs