agentifind-benchmark

To benchmark AI agent performance with codebase documentation, this skill automates setup by creating hook scripts and running parallel agent tests to directly compare efficiency and accuracy with and without the CODEBASE.md guide. It logs tool calls and violations for analysis.

When testing AI agent effectiveness on a codebase, metrics captured include tool call counts, file access patterns, and agent turns. The benchmark tracks how quickly and accurately agents complete representative coding tasks with and without the documentation guide.

Yes, you need to generate a codebase.json file and CODEBASE.md prior to running AI agent benchmarks. This skill requires the prior execution of an agentifind tool to produce these necessary input files for setting up parallel agent tests and hook scripts.

Restricted file access violations are tracked during an AI agent benchmark using pre/post tool use hook scripts. The skill enforces guide restrictions and logs any attempts by the agent without the guide to access restricted files, demonstrating the documentation's protective role.

Yes, you can run parallel agent tests for both application and infrastructure code. The skill generates benchmark templates tailored for either application or infrastructure code, allowing you to measure documentation effectiveness across different codebase types.