Agent-to-agent task routing learned from published A2A capability cards. Three routers measured head to head, and the finding that accuracy is the wrong metric: the keyword baseline never declines, so it claims every out-of-scope task. PyTorch, scikit-learn, sentence-transformers.
Your prompt says 'don't mention billing.' Your agent mentions billing. Pink Elephant finds the don'ts in prompts, skills, CLAUDE.md and AGENTS.md, rewrites them, and checks the rewrite with a blind reader. 92% fewer mentions.
The open source AI engineering platform for agents, LLMs, and ML models. MLflow enables teams of all sizes to debug, evaluate, monitor, and optimize production-quality AI applications while controlling costs and managing access to models and data.
A local, Linux-first profiler for the inner developer loop: sample a command's whole process subtree from /proc on a fixed cadence and write a deterministic, diffable bundle.