deepagents

Test code built on AI SDK v7 using its real `ai/test` utilities and V4 provider protocol. Use whenever writing, fixing, reviewing, or migrating tests around AI SDK agents, `generateText`, `streamText`, structured output, tools, embeddings, image/speech/transcription/reranking/video models, provider registries, or simulated streams. Trigger for phrases such as "test my agent", "mock the model", "fake the stream", "simulate a tool call", "test retry logic", "inspect the system prompt", "my streaming test hangs", or "empty result.text", even when the user does not explicitly name AI SDK. This skill is v7-only: remove V3 fixtures and custom helpers that duplicate `ai/test`.

Analyze a target piece of code and propose how to make it reusable across the rest of the codebase. Walks through phases — understand → survey → design → migration plan — and only edits code after the user approves the plan. Codebase-agnostic. Run only when the user explicitly invokes it (e.g. `/reuse`, "make this reusable", "extract this", "find duplicates of this").

Update the Text2SQL system prompt in a generalizable, non-overfit way using failure-pattern driven evaluation loops. Use when SQL generation quality regresses or specific failure classes repeat.

Write conventional commit messages with type, scope, and subject when the user wants to commit changes or save work.

Generate documentation for a package by analyzing source code. Use when the user wants to write, create, or generate new documentation pages for a package from scratch.

Update documentation based on code changes since last push. Use when the user wants to update, sync, or refresh documentation to match recent code changes across packages.

Keep repository documentation accurate and agent-legible through drift detection and deterministic fixes. Use when asked to scan docs for stale content, broken links, metadata/nav drift, API coverage gaps, or to run recurring documentation gardening in on-demand or scheduled maintenance loops.