What problem does it solve? Persistent data outlives every process that writes it, yet storage risks like non-atomic writes, unversioned formats, missing migrations, and unbounded accumulation are hard to spot in a large codebase. This Skill reads a project's persistence layer and produces verifiable findings about where data lives, how it is written, and what happens when a writer crashes or formats change. ## Core Features & Use Cases - Data-class inventory: Maps every kind of persistent data (config, state, caches, databases, outputs) with its location, format, owner, writer count, and sensitivity. - Mechanism analysis: Examines access patterns (ORM vs raw SQL, transactions, locking, streaming vs whole-file), schema ownership and migrations, format versioning and compatibility, atomic writes and corruption recovery, and lifecycle (retention, cleanup, backup, export). - Deterministic counting script: Ships a standard-library-only Python script that counts file I/O, serialization, SQL, migration, checksum, and locking sites per ecosystem, excluding test code, to ground findings in facts. - Use Case: Point it at a desktop CLI app with a home directory to learn that session transcripts are append-only JSONL with no version field, that config writes are non-atomic, and which data class is most at risk. ## Quick Start Ask the AI to run the storage scan on this repository and report where the project stores its data and which data class is most at risk.
