Devzy v2.3.0 Release Notes

This release delivers a major data layer modernization, enterprise source control support, significant review quality and context handling improvements, and comprehensive security hardening including browser and network policy improvements.

New Features

GitHub Enterprise Support

Full support for GitHub Enterprise Server deployments.

• Configurable API base URL: Supports custom enterprise endpoints
• URL validation and normalization: Ensures correct API base URL formatting
• Enterprise-compatible early review indicator: In-progress status comments work with Enterprise installations

Python Import Bindings in Parser

• Implemented import binding extraction for Python code
• Improved semantic understanding of Python dependency graphs
• Better context retrieval for Python codebase reviews

Improvements

Architect Review Quality

• Improved groundedness and reduced false positives: Refined architect-level review prompts and validation
• Claim verification enhancements: Bounded concurrency for processing claims, preventing overload
• Discarded invalid claims: Reduced misattribution risks in claim extraction
• Mutual exclusivity: Enforced between verified and contradicted claims
• Judge prompt refinements: Clarified scoring criteria and handling of contradictions

Code Graph & PR Review Isolation

• Prevent PR pollution: PR reviews no longer pollute the persistent code graph
• Optional lazy-initialized extractor: Improved handling when parsing fails
• Preserved layer violations: Persisted violations remain when analysis fails

Review Context Handling

• File-specific context filtering: Context is filtered per file for more relevant results
• Priority-based token budget allocation: Modular logic for allocating tokens across review context
• Consistent ranking: Fixed similarity vs ranking score confusion for context quality
• Unclosed fence detection: Improved handling of code blocks in review context
• Centralized embedding thresholds: Reduced magic numbers across the codebase

Language-Agnostic Improvements

• Language-specific issue detection improvements across Go, Python, and Java
• Better handling of parser edge cases

Bug Fixes

• Fixed: Credential errors on long pull request reviews via improved auth strategy
• Fixed: Invitation API failure by bumping neatcode-data version
• Fixed: Invitee GitHub username incorrectly stored in invitation data
• Fixed: Retention days type (integer instead of bigint) in data layer

Infrastructure

Data Layer Migration

Major data layer migration to a more consistent relational data architecture.

• Shared data package: Migrated repositories, invitations, and core entities to the new schema
• Row-level isolation: Stronger tenant data boundaries at the database layer
• Backend & Frontend: Full integration with the updated data client
• Runtime packaging: Added deployment-compatible binary target handling

Security Hardening

• Nonce-based CSP: Implemented Content Security Policy nonce handling to mitigate script-injection risks
• CORS auth bypass fix: Resolved authentication bypass vulnerability
• Framework update: Updated frontend framework version to resolve security vulnerabilities
• Dependency updates: Resolved security vulnerabilities across core dependencies
• Session invalidation: User session invalidated immediately when account is disabled

Testing & CI

• Integration tests: Enhanced GraphReviewIntegration and related test coverage
• Jest configuration: Restructured for isolated testing of Go and Python parsers
• CI workflow: Improved integration test execution

UI & UX

• Repository table: Replaced "-" placeholders with clearer empty states
• Custom typography: Added WOFF2 font support, typography variables, and letter/word spacing