Questionnaire stage skill (detect + gather)

[savitharaghunathan]

Part of #54.

Questionnaire

Combines automated detection with decision gathering in a single stage.

Consumes	Source repo
Produces	.konveyor/questionnaire.json, .konveyor/results.json entry
Image	TBD (may reuse agent-plan if graphify lands here)

Two phases

1. Detect — analyze the project and produce a tech summary (languages, frameworks, build tools, app servers), architecture summary (layers, complexity), and dependency capture
2. Gather — collect migration decisions: target framework, scope constraints, priorities

Graphify placement (here vs plan stage) is TBD.

Interactive vs non-interactive

• Interactive: agent presents detection results, asks the human for decisions
• Non-interactive: LLM picks the best answers based on detection results and records its reasoning in questionnaire.json. Eval stage later scores whether those choices led to a good migration.

Expected artifact

.konveyor/questionnaire.json:

{
  "detection": {
    "languages": ["java"],
    "frameworks": ["javaee-7", "weblogic"],
    "buildTool": "maven",
    "appServer": "weblogic-12c",
    "sourceFiles": 142,
    "archSummary": "3-tier: entities → services → REST controllers",
    "dependencies": ["javax.ejb", "javax.persistence", "javax.jms"]
  },
  "decisions": {
    "targetFramework": "quarkus-3",
    "scope": "full",
    "priorities": ["correctness", "minimal-diff"],
    "constraints": ["preserve-test-structure", "java-17-minimum"]
  },
  "mode": "non-interactive",
  "reasoning": "Selected quarkus-3 based on detected JPA + JAX-RS usage and Maven build"
}

Tasks

• Define questionnaire.json schema
• Create skills/questionnaire/SKILL.md
• Decide graphify placement (questionnaire vs plan)
• Define how interactive input reaches the agent
• Record LLM reasoning for non-interactive mode (consumed by eval stage)