What problem does it solve? Large AI-assisted code migrations fail when agents produce plausible-looking code without objective verification. This Skill turns a migration into a controlled operating model — a rulebook, work queues, reviewer prompts, and a mechanical parity judge — so behavioral equivalence with the old system is proven, not assumed. ## Core Features & Use Cases - Candidate Filter: Walks a go/no-go decision table (measurable pain, capturable behavior, bounded scope, cheap rollback) and recommends precursor work instead of migrating when the filter fails. - Judge-First Verification: Builds golden scenarios from the old system and wires them to a shipped parity differ (parity_diff.py) that normalizes JSON outputs, ignores noise keys, tolerates float drift, and must be proven to catch a seeded break. - Lintable Migration Control Pack: Authors eight artifacts (RULEBOOK.md, DEPENDENCY_MAP.md, GAP_INVENTORY.md, PORTABILITY_TEST_PLAN.md, parity_check.*, AGENT_WORK_QUEUE.md, REVIEWER_PROMPTS.md, PHASE_GATES.md) and lints them with control_pack_lint.py until PACK_RESULT: PASS. - Four Standalone Modes: Run the full workflow or invoke economics, judge, pack, or qualify independently. - Use Case: Porting a Python payments service to TypeScript — qualify the candidate, capture golden repayment-schedule scenarios, lint the control pack, pilot a disposable slice, then fan out implementer and reviewer agents under phase gates. ## Quick Start Ask the agent to run the ai-migration-operating-model skill to qualify and plan migrating your service from Python to TypeScript.
