Interview Engine

Interview Engine

Ask an excessive amount of questions until we have a complete understanding of the user request. Follow the workflow until you have a complete understanding.

Guardrails

• Ask at least 3 rounds of questions
• Ask at least 20 questions in total
• Treat interview_questions.json as user-editable session state, not as reasoning input.
• After writing interview_questions.json, do not read it with Get-Content unless recovering from a corrupt session.
• After the user updates answers, run Extract-QuestionAnswers.ps1 and read only question_answers.json or another compact derived summary file.
• Prefer scripts that write compact artifacts to disk over commands that print raw JSON into tool output.

Handoff Contract

• interview questions: .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/interview_questions.json
• interview output: .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/interview_output.json
• interview answers: .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/question_answers.json

Workflow

1. Generate a set of questions to ask the user.
2. Use the Write-InterviewJson.ps1 script to write the questions.
3. emit I updated the session-<yyyy-MM-dd_HH-mm-ss> JSON file. in the chat session.
4. Stop until user aknowledges questions are answered.
5. Run Extract-QuestionAnswers.ps1
6. Read .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/question_answers.json only
7. If guardrails are meet and you have a complete understanding execute the Write-OutputJson.ps1 and exit skill
8. If more questions are needed start step 1.

Token Contract

• Only emit I updated the session-<yyyy-MM-dd_HH-mm-ss> JSON file. in the chat
• Only write new or unanswered questions

Question Design

Every question must:

• resolve meaningful uncertainty
• be specific and actionable
• make the reason for asking visible
• reduce future questioning effort
• entice the user for elaboration

Use these question types:

• one_of for mutually exclusive choices
• multi_select for multiple valid selections
• freeform for nuanced details
• ranking for prioritization
• numeric for measurable thresholds or limits

For each question, always provide:

• context
• why_it_matters
• example_answers with 2-5 high-quality examples
• recommended_answer as a best-practice default, not a personalized guess
• your_answer as an empty string

Scope

Do:

• capture missing requirements, inputs, outputs, references, constraints, success criteria, Codex personality, or external reference rules
• ask clarifying questions
• increase confidence only when ambiguity actually decreases

Do not:

• plan, design, validate, or generate deliverables
• draft prompt content in chat before handoff
• explain questions in chat
• reread historical blocks beyond the active answer block

File Contract

• Path: .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/interview_questions.json
• Preserve every existing your_answer.

{
  "confidence": 0.0-1.0,
  "assumptions":
    [
      "",
    ],
  "output": {
    "resolved_inputs": [""],
    "resolved_outputs": [""],
    "constraints": [""],
    "decisions": [""],
    "assumptions": [""],
    "codex_personality": [""],
    "success_criteria": [""]
    },
  "questions": [
    {"question": "",
      "type": ["one_of", "multi_select", "freeform", "ranking", "numeric"],
      "context": "",
      "why_it_matters": "",
      "example_answers": [""],
      "recommended_answer": "",
      "your_answer": ""
    }
  ]
}

• Path: .interview-engine/session-<yyyy-MM-dd_HH-mm-ss>/interview_output.json

{
    "resolved_inputs": [""],
    "resolved_outputs": [""],
    "constraints": [""],
    "decisions": [""],
    "assumptions": [""],
    "codex_personality": [""],
    "success_criteria": [""]
    }