Feature Readiness PREP Agent on DAP

Create an AI-powered agent using GitLab's Duo Agentic Platform to streamline the Platform Readiness Enablement Process (PREP) by automatically collecting evidence from GitLab artifacts, intelligently selecting relevant questions, and generating evidence-based assessment responses.

Problem Statement

The Platform Readiness Enablement Process (PREP) is essential for ensuring features are production-ready across GitLab.com, GitLab Dedicated, and Self-Managed platforms. However, the current manual process creates significant friction:

For Feature Teams:

  • Time-intensive process: Teams spend days to weeks navigating comprehensive checklists across 11 categories
  • Redundant documentation: Information already documented in issues, MRs, and design docs must be manually extracted and reformatted
  • Cognitive overload: Determining which questions are relevant from hundreds of possibilities requires deep understanding of infrastructure concerns

For Platform/Reviewing Teams:

  • Review bottleneck: Increasing number of features requiring simultaneous assessment creates unsustainable reviewer workload
  • Validation overhead: Manually verifying evidence links and technical claims across multiple assessments

Solution Overview

An intelligent PREP assistant that transforms the assessment from a manual checklist into an automated, conversational process.

Key Capabilities

Automated Evidence Collection

  • Scan project repositories, issues, MRs, and wikis for relevant documentation within GitLab
  • Map discovered evidence to specific PREP questionnaire requirements
  • Identify gaps where additional evidence is needed

Intelligent Question Selection

  • Analyze feature characteristics to determine applicable PREP categories
  • Start with foundational questions (architecture, security, platform strategy)
  • Progressively expand to additional categories based on feature maturity
  • Skip irrelevant questions based on feature scope and deployment targets

Evidence-Based Answer Generation

  • Generate draft answers using collected evidence
  • Provide confidence scores for each response
  • Include direct links to supporting documentation
  • Highlight areas needing human expertise or clarification

Human-in-the-Loop Refinement

  • Present high-confidence answers for batch approval
  • Flag low-confidence areas for focused human review
  • Request specific missing information with clear guidance
  • Enable iterative refinement through conversational interaction

Phased POC Breakdown for PREP Agent

Edited by Sofia Vistas