vision-alignment-foresight

Restate the durable intent separately from the implementation, map actual behavior from docs, code, and config, then grade evidence as observed, researched, inferred, or speculative. The Skill outputs an alignment verdict, critical gaps, and a keep, refine, pivot, or kill recommendation.

Run the critical lenses covering product, engineering, agent experience, maintenance, ecosystem, and adversarial perspectives. Then forecast across now, 3-6 months, and 12-24 months with confidence levels, assumptions, and falsifiers that would trigger a pivot.

Use it when the decision is not yet made and you need to test whether a vision is worth building. Once a decision is accepted, route to repository-governance-lifecycle to formalize it as an ADR instead.

Concrete signals work best: usage logs, eval results, issue and PR history, support questions, benchmark outcomes, adoption friction, failing tests, and rollback history. Future claims should be researched from primary sources and tied to a decision or falsifier.

It does not replace formal decision records, broad multi-agent critique, or skill quality measurement. Forecasts are inherently uncertain, so every prediction must state a time horizon, confidence, assumptions, and a falsifier rather than claiming certainty.