agent-audit v0.12.0 - Semantic Layer Refinement

🎯 Semantic Layer Refinement

This release completes the semantic layer refactoring for AGENT-034 and AGENT-018, achieving 100% false positive elimination in benchmark projects.

Core Principle

"Identify what to check (finite), not what to exclude (infinite)"

What's New in v0.12.0

🔧 Insert Ambiguity Fix

• list.insert() no longer triggers AGENT-018 false positives
• index.insert() with llama_index import still correctly triggers

🚀 OpenAI/Anthropic Function Calling Detection

• Functions registered via tools=[func1, func2] now detected as Tool entry points
• Enables AGENT-034 detection for non-@tool frameworks

Benchmark Results

Scenario	Result
SWE-agent AGENT-034 FP	33 → 0 ✅
Generative Agents AGENT-018 FP	11 → 0 ✅
Test Suite	1039 tests passing