interviewer

Interviewer — Active Workflow Discovery

This is the core of workflow-architect's active interrogation mode. It runs a structured but adaptive conversation with the user to discover how they work.

State Model

The interview builds a structured representation of the user's workflow. State is stored via the memory tool with the prefix workflow-architect:state: so it survives across turns.

state:
  entry_points: []          # How sessions start
  phases: []                # Distinct phases discovered
    - name: string
      description: string
      typical_tools: []
      typical_openers: []   # What user says to enter this phase
      typical_exits: []     # What user says to leave this phase
  branching: []             # Decisions and what drives them
  pain_points: []
  exit_criteria: []         # How sessions end
  archetype: null           # Best match from workflow-archetypes
  convergence_score: 0      # 0.0 to 1.0 — enough to generate bundle?

Question Progression

The interview follows a branching script. Each answer feeds the state model and determines the next probe. Do not ask all questions sequentially — adapt based on what the user has already told you.

Phase 1: Session Opener (1-2 questions)

Start broad. The goal is to understand the user's self-model of their workflow.

Agent: "Walk me through a typical session from the very start.
        What's the first thing you do when you open this agent?"

User:  "I usually check my task list, see what's urgent, and jump into the
        most pressing issue."

Agent: [Records entry point: "check task list, prioritize by urgency"]
       [Probes for structure: "After that initial triage — what happens
        next? Does the session settle into a rhythm?"]

Alternative openers (pick one based on the user's stated context):

• "Describe a session that went really well. What did it look like from start to finish?"
• "What does a typical day look like, broken into sessions?"
• "If I looked at your last 10 sessions, what patterns would I see?"