Rohan Rao

You can design a memory architecture from your actual codebase, compare AI memory frameworks head-to-head, audit the memory landscape every 2-3 months against your existing architecture doc, and deploy the approved design as a production-ready memory system with enforced quality checkpoints.

Engineers building AI applications that need persistent memory should use them. memory-design suits architects evaluating stacks via codebase analysis and live research; memory-ship suits developers ready to implement, since it refuses completion until every build, hardening, and eval check passes.

Each skill is triggered by a slash command: /memory-design to produce the architecture doc, /memory-compare for framework evaluation, /memory-audit for periodic re-research, and /memory-ship to deploy. memory-ship requires memory-design/architecture.md to exist before it can run its 11-phase build.

memory-ship requires an existing memory-design/architecture.md produced by the memory-design skill. It first reads your project, then executes an 11-phase pipeline covering prerequisites, setup, wiring, quality, hardening, and evaluation, refusing to declare done until every check passes.

The memory-audit skill should be run every 2-3 months. It re-researches the current AI memory landscape and diffs the findings against your existing architecture document, keeping your memory stack recommendation aligned with newly released frameworks and tooling changes.