knowie-migrate

What problem does it solve? Long-running projects lose the reasoning behind their decisions: the knowledge base drifts out of date, old structures no longer match current conventions, and the 'why' behind past choices is buried in git history. This Skill rebuilds a project's knowledge base by replaying its git history, recovering each decision's rationale without contaminating it with hindsight. ## Core Features & Use Cases - Encoder-decoder reconstruction: A bidirectional pass over the whole git history emits only structure (slices, transitions, tombstones), then parallel causal-masked sub-agents recover each slice's 'why' blind to its own future. - Transition vs tombstone classification: Distinguishes decisions that were in force then changed (numbered history entries) from options considered and declined (grouped tombstones) and knowledge-only projection edits (no transition). - Rumination loop: Re-reads the assembled base against git as fixed evidence, running judge passes and human-in-the-loop denoising until the structure converges. - Use Case: A team adopting a new knowledge-base structure runs this Skill on their repository; it walks first-parent merges, replays each era as the developer would have recorded it, and produces a migrated base with dated history entries, episodes, and drafts in the project's original language. ## Quick Start Ask the AI to run the knowie-migrate skill to check this knowledge base for structure drift and rebuild it from git history.