LLM-based behavioral analysis for the semver-analyzer.
This crate implements the BehaviorAnalyzer trait from semver-analyzer-core.
It provides:
- Agent-agnostic LLM invocation via
--llm-command(goose, opencode, etc.) - Template-constrained spec inference — prompts that produce
FunctionSpecJSON - Tier 1 structural spec comparison — mechanical comparison without LLM
- Tier 2 LLM fallback — for ambiguous
notesdiffs and fuzzy matches
Usage
use LlmBehaviorAnalyzer;
let analyzer = new;
let spec = analyzer.infer_spec?;