How It Works

GitSniff's code review follows a sophisticated multi-step pipeline designed to provide thorough, accurate feedback.

Step 1: Pre-analysis

When a PR is created or updated, a fast AI model quickly scans the changes to:

• Assess the overall risk level of the changes
• Identify which types of issues might be present
• Recommend which security scanning tools should run
• Determine the optimal model for deep analysis

This step uses a lightweight model to keep costs low and response times fast.

Step 2: Security scanning

Based on pre-analysis recommendations, GitSniff runs the appropriate security tools in parallel:

Only relevant tools are run based on the changes in the PR, keeping scan times minimal.

Step 3: AI analysis

The selected AI model performs a comprehensive review that considers:

• The code diff (added, modified, and removed lines)
• Security scan results from Step 2
• Repository context (if enabled)
• Custom instructions (if configured)
• The programming language and framework being used

The model produces:

• A risk score (0-100)
• Categorized findings (bugs, security, performance, style)
• Suggested improvements with code examples
• An overall assessment

Step 4: Results delivery

GitSniff delivers the results back to your pull request:

• Inline comments on specific lines of code
• PR summary in the pull request description
• Status check (clean, issues, or critical)
• Review comment with the full analysis

Model selection

GitSniff can automatically select the best model based on PR size:

You can override this with manual model selection or configure size-based preferences in your settings.