Today we’re extending the same model to Spec Reviewer. Unmet requirements now appear inline, pinned to the exact code paths that cause them, and reviews include screenshots from the agent’s sandbox session showing the evaluated behavior.
Unmet requirements appear inline as embedded review suggestions, with sandbox screenshots showing exactly how behavior was evaluated.
When Spec Reviewer evaluates a change, it no longer reports unmet requirements as detached feedback. Each unmet recommendation is now embedded inline, pinned to the exact lines of code responsible for the issue.
The reviewer explains why the requirement is unmet, how it affects behavior, and what prevents the system from recognizing the integration or flow as valid. This removes the need for reviewers or authors to infer where a requirement applies.
For behavior-dependent checks, Spec Reviewer also captures screenshots from the agent’s sandbox session and attaches them directly to the review comment. Reviewers can see the activation dialog, integration state, or UI flow exactly as the agent observed it during evaluation.
The result is a review comment that contains the requirement, the precise code location, and visual evidence of the evaluated behavior in one place.
Inline requirement violation, pinned to code
Spec Reviewer flags the unmet Shortcut integration requirement directly on the code path responsible for activation state tracking, explaining why the integration can never be marked active.
Behavior evidence from sandbox session
The review includes screenshots from the agent’s sandbox session, showing the activation dialog and integration state that caused the requirement to fail.
Configuration and metadata mismatch surfaced inline
Spec Reviewer highlights where integration metadata is missing from the workspace payload, explaining why the UI cannot display status badges even when the integration is connected.
.png)