rad

Run the verification workflow with the changed-files scope to scan git-modified files for assumption tags. Critical assumptions route to premium models while standard ones use free models, and the report lists blocking issues with suggested defensive code fixes.

Add comment tags like #CRITICAL, #ASSUME, or #EDGE followed by a category and description, plus a #VERIFY line describing the defensive code needed. Only tag assumptions that could cause production failures, not trivial logic.

#CRITICAL marks assumptions that could cause outages or data loss, such as payments and concurrency, and routes to premium models. #ASSUME covers standard bug risks verified with free models, while #EDGE covers rare scenarios handled by fast models.

Multi-model verification requires OPENROUTER_API_KEY to be set. Without it, the workflow degrades to single-model verification and tags output as VERIFIED-SINGLE-MODEL so readers know decorrelation was not achieved.

The same context that made an assumption cannot effectively review it due to confirmation bias. Routing each assumption to a fresh model context provides independent analysis that catches blind spots the original author missed.

Do not tag trivial or obviously safe logic, such as basic arithmetic, since over-tagging creates noise. Reserve tags for production-impacting assumptions involving timing, external resources, data integrity, concurrency, security, or payments.