Mission Run Overview
Mission Run: Retrospective Learning Default-On Policy
Make retrospective learning a default-on, policy-driven feedback loop with first-class authoring commands, replacing env-var gating.
Today retrospective learning hides behind SPEC_KITTY_RETROSPECTIVE and SPEC_KITTY_MODE env vars, the runtime passes facilitator_callback=None so the strict path fails closed, and agent retrospect synthesize is the de facto authoring surface despite being a proposal preview/apply command. As a result, completed missions rarely produce useful retrospective.yaml records, and PR #1136 docs now overstate what summary/synthesize actually do. This mission lands a durable RetrospectivePolicy resolver sourced from .kittify/config.yaml and charter, a real generator wired into the runtime, post-completion best-effort behavior by default with opt-in strict pre-completion gates, first-class retrospect create and backfill CLI surfaces, env-var deprecation as test/dev overrides only, and docs/skills cleanup so the 3.2.0 retrospective learning story is coherent end-to-end.