Index
agent_cover
AgentCover: Logical coverage metrics for LLM Agents.
This library provides a comprehensive toolkit to measure how well your test suites exercise an LLM agent's capabilities. Unlike traditional code coverage, AgentCover focuses on:
- Prompt Coverage: Ensuring all templates and raw string prompts are used.
- Tool Coverage: Verifying that the agent effectively invokes its available tools.
- Decision Coverage: Validating that the LLM's outputs hit all expected business logic branches (e.g., specific intents or status codes).
The library utilizes a hybrid instrumentation strategy, combining static analysis of the codebase with runtime monkey-patching of popular frameworks like LangChain, LlamaIndex, and PromptFlow.