v3.2.4

[3.2.4] - 2026-02-08

Changed

• Refactored agent analyze and review commands from parallel execution to a pipeline architecture where static analysis is source of truth and LLM verifies
• Extracted orchestration logic from cli.py into pipeline.py
• Static analysis now indexes the full project even when reviewing changed files, fixing FP dead code in the review command
• LLM no longer independently discovers dead code. Only confirms static findings
• LLM-only findings are now always marked as needs_review and never block CI
• Removed dependency on line-proximity merger for combining static and LLM results
• Added node_modules to default exclude folders

Added

• New DeadCodeVerifierAgent that reviews static findings with call graph evidence and defs_map context
• New pipeline.py containing run_pipeline and run_static_on_files
• Added unit tests for pipeline.py covering static analysis categorisation, LLM dead-code verification, LLM-native analysis, deduplication, sorting, and fallback behaviour
• Added unit tests for dead_code_verifier.py covering verdict application, confidence parsing, context building, single and batch verification, annotation, and agent initialisation

Fixed

• Fixed run_gate_interaction because it uses * to enforce keyword only args
• Fixed circular dependency checker feeding .ts/.go files to ast.parse()
• Fixed undefined task variable in run_pipeline progress spinner