Add YYLO Benchmark to LLM Evaluation Tools - #217

YYLO Benchmark is a CLI for longitudinal evaluation of agent runs — task prompts execute in private fresh-repository attempt workspaces under deterministic commands and configurable LLM judges, with hash-linked receipts, manifests, and provenance retained as immutable evidence.

• Two deliberately separate evaluation lanes: isolated v2 (default — private fresh-repository attempt workspaces) and an explicitly authorized governed lane with blinded per-step judging
• Evaluator profiles support deterministic commands and configurable LLM judges; a required deterministic correctness or safety failure cannot be overridden by judge prose
• Installable via npm as @yylo/benchmark (187 downloads last month); the companion @yylo/cli coding-agent orchestrator is at 794/month
• Open source (MIT), pushed near-daily since January 2026

Sits next to DeepEval / InspectAI / EvalView as an evaluation-layer tool for agents — not a prompt library.

Disclosure: I am on the YYLO team; this PR was prepared with AI assistance. Happy to revise, trim, or re-file if the placement or format needs adjusting.