← Back to Portfolio
Pinecone · project

Cultivar — test your agent skills

A CLI from the Pinecone DevRel team for writing tests for agent skills and benchmarking them across Claude, Copilot, and Gemini — locally or in sandboxes.

Cultivar is an open-source CLI from the Pinecone DevRel team for testing the skills you write for AI agents. It lets you define tasks, run them across multiple agents, and grade the results — so you can tell whether a skill actually helps before you ship it.

What it does

  • Benchmarks skills across Claude, Copilot, and Gemini side by side, with built-in with-skill / without-skill / with-docs variants
  • Runs evaluations locally or in isolated Modal sandboxes for parallel testing at scale
  • Grades task success automatically, with criteria-based feedback
  • Surfaces detailed traces, transcripts, token counts, and cost per run
  • Scaffolds new tasks and skill definitions with a single command

Built in Python on top of Modal, integrating the Anthropic, OpenAI, and Google agent CLIs.

Skills & Technologies

Python Agent Skills Evals Modal CLI Testing Claude Gemini
// stay in touch

Get Answering Machines in your inbox.

Occasional notes on AI, coding agents, and where this is all headed — plus the very occasional origami diagram.

Arjun Kirti Patel
Chicago, IL · folded with care
© 2026 Arjun Kirti Patel · MIT Licensed