1. Task setup
Confirms the upstream function you want to accelerate and scaffolds a fresh task directory (task.yaml, pipeline/, reference.{py,R}, evaluate.{py,R}, data/).
No separate install step. The prompt’s first action is to check for the zyme CLI and install it if missing, then scaffold your task. Just paste it. Prefer to set up the toolchain yourself first? See Environment setup.
Open a fresh coding-agent session in an empty workspace directory, then paste:
Pick Bio if your target is single-cell / genomics / Bioconductor / scanpy / Seurat. Pick OtherField for astronomy, climate, chemistry, physics, statistics, anything else.
Time: ~2 minutes.
What to watch for: a new <task_name>/ directory with task.yaml filled in, and a Yes, ready to proceed from the agent. If the agent asks clarifying questions, answer them: it needs to lock the function signature and the concordance gate.
Next -> 2. Benchmark initialization.