a small sandbox for testing multi-step AI prompt chains visually.

why — i wanted to see how prompts compose before wiring them into production.

learned — sequential chains are fragile. branching and fallback paths matter more than the happy path.

a tiny CLI that reads a YAML task file and runs steps with retries and logging.

why — i kept writing the same bash scripts. this replaces them with something testable.

learned — YAML is fine for config but terrible for logic. keep task definitions declarative.