Plan stage skill

[savitharaghunathan]

Part of #54.

Plan

Reads the project and questionnaire output, produces a migration spec for user approval, then generates the implementation plan.

Consumes	Source repo, .konveyor/questionnaire.json, domain skill references (/opt/skills/*/references/)
Produces	SPEC.md (migration spec for approval), PLAN.md (implementation plan), graph.json (if graphify runs here), .konveyor/results.json entry
Image	agent-plan (Python 3, graphify)
Mode	Interactive (spec approval) + non-interactive (plan generation)

Migration intelligence (rules, extracted patterns, solution server) enters via domain skill references. Graphify placement (here vs questionnaire stage) is TBD.

Two phases

1. Spec — produce SPEC.md summarizing what will be migrated, the approach, and key decisions. In interactive mode, the user approves or rejects. In non-interactive mode, the LLM approves its own spec and records reasoning.
2. Plan — once the spec is approved, generate PLAN.md with step-by-step implementation details.

Expected artifacts

SPEC.md — migration scope, approach, key decisions, risk areas. Approved or rejected by user (or LLM in non-interactive mode).

PLAN.md — one step per file, exact paths from graph.json, dependency order, COMPLEX-flagged steps for structural changes. Verification section with build/test commands.

graph.json — code graph with communities, god nodes, dependency edges.

Tasks

• Integrate questionnaire.json as plan input
• Define SPEC.md structure and approval flow
• Validate PLAN.md structure contract (one file per step, dependency order, layer order)
• Ensure domain skill references are consumed for migration patterns
• Skill-level chunking strategy for large projects (50+ plan steps)