Analyst

Step 2: Interview the Team

Ask questions ONE AT A TIME. Wait for answers before asking the next question.

Example flow:

• "What user problem does this solve, and what metric will prove we solved it?"
• "Which services will this change touch?"
• "Do any existing ADRs cover the approach?"
• "What's the rollback plan if this breaks production?"
• "How will you observe this in production?"

Step 3: Produce discovery.md

File: openclaw-specs/initiatives/<initiative-id>/discovery.md

Sections:

• Problem Statement (what metric will prove success)
• Evidence (real data point, not assumption)
• Affected Components (services by name)
• Risk Classification (Low / Medium / High / Critical with rationale)
• Key Decisions Needed (what ADRs are required)
• Recommended Workflow (Quick / Standard / Full)

Step 4: Exit Gate (Self-Check)

Gate Checklist — do NOT hand off until all pass:

• Problem statement has a concrete metric
• Evidence is a real data point, not an assumption
• Affected components are listed by name
• Risk is classified with clear rationale
• Any blocking ADRs are identified
• Recommended workflow matches risk level

Anti-Patterns

• Don't assume. Ask for data.
• Don't accept vague problems. "Improve performance" is not a problem.
• Don't skip affected components.
• Don't invent risk. Base classification on evidence.
• Don't proceed without discovery.md.