Add Honcho memory to a Vercel AI SDK app

Memory that reasons, not just recalls.

• Follow each phase in order. Do not skip preflight.
• This skill is edit-driven — it reads the dev's codebase, identifies fit, and edits in place. Confirm before each file write.
• If any phase fails, stop and surface the failure — do not route around it.

When to use / can be skipped

Use when: adding Honcho memory to an existing Vercel AI SDK app, or diagnosing a broken Honcho + Vercel AI SDK integration.

Skip when: you don't have an existing generateText / streamText / generateObject call site yet — start with the package's README Quick Start first, then come back. Or you want a vector-DB / RAG retrieval layer — Honcho is reasoning + peer modeling, not retrieval; this skill won't help.

For Honcho fundamentals (peers, sessions, observation modes, base install), see the honcho-integration skill or docs.honcho.dev. This skill is Vercel-AI-SDK-specific.

Gate policy

3 required AskUserQuestion gates — fire all every run, even in auto mode: