Structured interview technique for process discovery using BPM/BPA methods, inverse engineering on vague descriptions, and organized question blocks. Use when age-spe-process-discovery needs to elicit complete, unambiguous process information from the user through a structured dialogue.
Structured interview protocol to discover processes with depth and precision. Combines BPM, BPA, and reverse engineering techniques.
Input:
express or architectOutput:
Before formulating any question, analyze the initial description and identify:
Prioritize questions about what is missing. Do not ask what you already know.
One question at a time. Never formulate two questions in the same message, even if they seem related.
Validate before advancing. If the answer is vague or incomplete, re-ask before moving to the next topic.
Decompose the vague. If the user uses generic terms, decompose into specifics.
Do not assume. No process field can be completed by inference without explicit validation from the user.
The reverse engineering templates (to disambiguate terms), the Express Questionnaire, and the Complex Blocks for Architects no longer reside hardcoded in this Skill, but rather in a centralized resource to lighten the static memory load.
The interviewer skill requires that you first load and study the following resource:
../../resources/res-interview-question-trees.md
Extract from there the matrix corresponding to your operating mode (Express = Short Questionnaire, Architect = Blocks 1-6 with Dynamic Decomposition) and interview while respecting the cadence of one question per turn.
Before closing the interview in Architect Mode, actively validate understanding:
Step 1 — Process reflection: "Before continuing, I want to confirm I understood correctly. The process is: [summary in 3-5 steps]. Is that correct?"
Step 2 — Challenge with extreme cases: Once the flow is confirmed, ask at least 2 challenge questions about:
Example: "You mentioned the system sends a response to the client. What happens if the client doesn't respond within X time? Is there a retry or does it escalate to a human?"
Example 1 — Decomposition of vague description
User: "I want to agentize customer support."
Reverse engineering application:
Example 2 — Detection of hidden complexity
User: "It's simple, I just need it to classify emails and forward them to the correct department."
Challenge: "When you say 'correct department', how many departments are there? What happens if an email could go to more than one? And if the classification is not clear?"
→ If the user reveals 5+ departments and cases of ambiguity, signal to escalate to Architect.