Try DriftCheck on one workflow that could quietly regress
DriftCheck is an open-source, local-first npm package for reviewing AI behavior changes before they ship: prompts, agents, RAG, tool calls, and model migrations. Run it locally, inspect the report, then send the blunt feedback that would make it worth keeping in CI.
Run locally
npx @a2zai-ai/driftcheck init npx @a2zai-ai/driftcheck check npx @a2zai-ai/driftcheck diff
Nothing is uploaded by default. Publishing a proof card is optional and explicit.
Prompt and model changes
Compare baseline vs latest behavior when a prompt is rewritten or a model is swapped.
RAG and grounding
Check whether answers stay cited, scoped, and faithful after retrieval or chunking changes.
Agents and tool calls
Catch schema drift, skipped confirmations, weak recovery, and tool-routing regressions.
What to send back
The best feedback is specific. A real workflow, a real blocker, and a clear reason it would or would not fit your build process is more useful than a generic review.
- +What was confusing in setup or pack authoring?
- +Which starter pack would make this useful in your actual workflow?
- +What would make the output worth keeping in CI?
- +What proof card or PR summary would you share with your team?
Best first test
Pick one prompt, retrieval flow, tool-calling path, or model migration where a small change could pass normal tests but still break behavior.