What problem does it solve? Project knowledge scattered across a ki-search memory store is hard to query consistently: different memory types (strong relations, decisions, errors, pending changes) need different retrieval strategies, and query logic duplicated across many skills drifts out of sync. This Skill centralizes all ki memory lookups into one single source of truth with per-type strategies. ## Core Features & Use Cases - Unified query layer for seven memory types: topic memories, API/interface info, data flows, strong cross-module relations, decision records, error solutions, and pending-change ledger entries, each dispatched to its own reference strategy. - Directory-first navigation for enumerable types: uses ki_query_group two-level navigation (subtree then groups with mode="full") plus batch ki_get_module_info retrieval, avoiding semantic recall gaps for strong relations, decisions, and pending changes. - Dual-path retrieval with change context: when a diff is available (code review, bug impact analysis), runs directory filtering and semantic ki_search in parallel and merges results; degrades silently without blocking when ki is unavailable. - Use Case: Before modifying a payment module, ask what other modules are affected; the Skill lists all strong-relation entries for that module, retrieves the relevant ones in batch, and returns a coupling checklist with direction and strength. ## Quick Start Ask the AI to check what other modules are affected before changing the payment module, using the ki memory lookup.
