Stage 02 — Verify

Zero trust, multilayered verification. At every point in the loop.

Verification runs at two checkpoints: inside the agent’s inner loop as code is written, and at sandbox exit before the PR opens. At both points, SonarQube’s deterministic analysis and Gitar’s AI-native review run together. Each layer catches what the other misses.

Inner loop vs outer loop

Verification at two levels

Traditional CI verification waits until a pull request is opened. By that point the agent has already generated thousands of lines, made dozens of decisions, and created dependencies that are expensive to unwind.

AC/DC's Verify stage runs at two points: inside the agent's reasoning cycle (inner loop), and at sandbox exit before the PR is opened (outer loop). Issues are caught and blocked before they compound.

Sonar products in this stage
bash — run verification
# Inner loop — per file as agent generates sonar analyze --file <file> # Outer loop — at sandbox exit sonar auth status     # exits non-zero if gate fails # View all findings sonar list issues
What the quality gate checks

Deterministic-first analysis

- reliability, security, maintainability

- 40+ languages, zero false negatives on BLOCKER/CRITICAL

Gitar AI code review

- functional bugs, logic errors, behavioral issues

- context-aware, catches what deterministic analysis can’t reach

Architecture conformance

- current vs intended structure

- boundary violations and cross-module drift

When verification runs

Inner

Per file, per reasoning step. Sonar Vortex runs deterministic checks covering pattern matching, control flow, data flow, taint analysis, and SCA as the agent generates. Gitar reviews for functional correctness and logic errors in parallel.

Outer

At sandbox exit, before the PR opens. SonarQube runs its full analysis suite including architecture conformance. Gitar runs a final context-aware review for functional and behavioral issues. If either layer fails, the Solve stage triggers.

3.2%
Why verify in the loop

Only a small fraction of AI-generated code carries a real defect — but at agent scale that adds up fast. Verifying every change inside the loop catches those issues before they ever reach a pull request.

Products in the Verify stage

Deterministic-first, AI-augmented

Core

SonarQube

Deterministic-first analysis across 40+ languages. The quality gate for all code — human- or agent-written. Available as Server (self-managed) and Cloud.

Product page
Security

SonarQube Advanced Security

Deep security analysis with taint tracking, secret detection, and advanced SAST. Catches vulnerabilities that surface-level static analysis misses.

Product page
Inner Loop

Sonar Vortex

Runs Sonar’s full analysis engine inside the agent’s reasoning cycle — every file, every step. Same depth and precision as a full CI scan, delivered in seconds via a two-phase (CI capture, on-demand restore) approach.

GA Product page
Architecture

SonarQube Architecture

Verifies that agent-generated code conforms to your intended module structure. Catches boundary violations and cross-module drift before they reach production.

Product page
Ai code review

Gitar

AI code review that lives in your pull requests. Reviews code, diagnoses CI failures, generates fixes, and commits only when the build passes. Operates as an agent, not just a scanner.

NEW Product page
Integration

SonarQube MCP Server

Exposes all SonarQube capabilities as native MCP tools. Claude Code, Cursor, and any MCP-compatible agent can call architecture and quality tools natively.

Product page
IDE

SonarQube for IDE

Sonar analysis directly in VS Code, IntelliJ, and other IDEs. Developers and agents get real-time feedback on issues and coding standards as they write.

Product page
CLI

SonarQube CLI

Unified command-line interface for agentic workflows. Agents can run analysis, check quality gate status, and query findings directly from the terminal.

Product page