Baz Spec Reviewer: Bring Product Requirements Into Backend Review
Baz’s Spec Reviewer for Backend loads a running environment and verifies backend behavior against the constraints found in Jira tickets, technical requirements, and product specifications.
Backend quality rarely degrades because a line of code is incorrect. We have compilers, linters and AI code review tools to tell us when our code is incorrect. What actually breaks production is the cascading impact of small, valid changes interacting in complex ways across distributed systems.
Yesterday’s (November-18, 2025) Cloudflare outage is a textbook example: according to their own incident report, a legitimate database permissions change caused a ClickHouse query to return extra metadata, which doubled the size of a bot feature configuration file, exceeded an internal feature limit, triggered a panic in the core proxy, and ultimately caused 5xx errors across Cloudflare’s global network. No syntax bug or obvious logic error caused the outage; the software behaved exactly as written, but hidden assumptions about configuration size and data shape were violated once that small change propagated.
The challenge is that these regressions are visible in a diff so traditional code review cannot see these regressions. It cannot confirm paging, filtering, or contract alignment. Backend teams rely on late discovery and intuition.
Our recent release of Spec Reviewer was created to bridge this gap on the front end by bringing design intent and user stories directly into GitHub. Starting today Spec Reviewer now understands backend requirements the same way it understands design intent.
Baz’s Spec Reviewer for Backend loads a running environment and verifies backend behavior against the constraints found in Jira tickets, technical requirements, and product specifications. It’s available today as part of Baz’s Code Review Agent platform for Github and Gitlab.
A Specification-Aligned Approach to Backend Behavior
Spec Reviewer for Backend applies the same principle we introduced for front-end review: align code with the specification by observing real behavior, not by guessing from the diff. It reads backend requirements the way it reads design requirements, then verifies that the running system behaves according to those expectations.
With Spec Reviewer for Backend, teams get a deterministic validation step that ensures: Backend changes respect the constraints of the system
Product guarantees don’t degrade across releases
Data-dependent regressions are caught before they ship
Filtering, pagination, and query logic remain aligned with spec Behavior matches intent under real runtime conditions
This brings the same confidence we delivered to front-end design validation into the operational heart of the backend.
Under the Hood
1. It ingests the product specification - Structured inputs from Jira, system requirements, and data constraints are translated into behavioral invariants. These include pagination rules, data window limits, filtering semantics, response-shape expectations, and performance thresholds.
2. It runs the real system in an isolated environment - Using AWS AgentCore and Playwright MCP, Spec Reviewer loads the dashboard or feature inside a secure environment and interacts with it exactly as a QA engineer would. It triggers fetches, applies filters, inspects the DOM, and traces every network call.
3. It compares measured behavior to the specification - Spec Reviewer evaluates whether queries are bounded, whether pagination works end to end, whether filters run on the server, whether responses follow the expected contract, and whether performance stays within the defined limits. Any deviation appears inside the pull request as a structured finding backed by runtime evidence.
This process makes backend review empirical. Instead of relying on intuition or late discovery, reviewers see how the system behaves and whether that behavior matches what the product requires. Regressions that typically surface only after deployment become visible during review, and backend changes can be evaluated with confidence.
A Unified Review Layer for Both Front-End and Backend
Spec Reviewer now covers the full stack. It keeps front-end implementation aligned with design intent and ensures backend behavior respects the constraints that define how the system should operate. Product requirements, user expectations, and actual execution stay connected throughout the development process.