What problem does it solve? When asked to implement a work task, jumping straight into code without a plan leads to oversized PRs, missed edge cases, and lost specification context. This Skill structures the planning phase: it classifies changes into five PR-splitting types, derives expected behaviors from recorded decisions, and produces a plan.md (plus sub-plans) that must be user-approved before any implementation begins. ## Core Features & Use Cases - PR Splitting by 5 Types: Classifies changes into mechanical replacement, schema/data change, new module, behavior change, or removal, and splits work into reviewable PRs with branch strategy and dependencies. - Specification Derivation from cc-memory: Pulls decisions and precedents from the cc-memory database to derive assertable edge cases with explicit derivation memos linking business requirements to expected behavior. - Derivation Verification Sub-Agent: Launches a background verification agent that checks each planned behavior against decision text via verbatim quotation before user confirmation. - Use Case: A user asks to implement a [作業] task tracked in cc-memory. The Skill confirms scope with the user, produces plan.md with a PR table, GO-judgment predictions, and an edge-case coverage table, then waits for approval without writing any code. ## Quick Start Ask the AI to create an implementation plan for a work task that has no plan.md yet, and it will classify the changes, draft plan.md with sub-plans, and present it for your confirmation.