arenaworkflow★1
johnayoung/flywheel ↗What it does
Blind A/B eval of a prompt/skill change: run an old and a new variant on the same input, then diverse-
Best for
workflow automation
Inputs
- · workflow arguments
- · git repository/branch
Outputs
- · audit report
- · files/changes
- · plan document
Requires
- · git
- · Schema validation
Failure modes
- · Git conflict
Trust signals
- · Type-safe output schemas
- · 3-step process