executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
extension-and-evolutionExtend mergekit with custom merge methods and task graphs, or plan optional evolutionary merge searches with explicit dependency and evaluation gates.
Independent GitHub snapshot · Automated classification · Original documentation governs
Use this route for a custom merge method, a reusable task-graph component, or
mergekit-evolve. It is an extension/developer route, not the normal YAML
method-selection route.
Read custom-methods.md for the API patterns and troubleshooting.md for signature, registry, and task-graph failures.
Treat mergekit-evolve as optional and expensive. Preflight the evolve extra,
CMA-ES, evaluation backend, model/task config, storage, GPUs, and credentials
before constructing a run. The inspected environment intentionally omits
ray, cma, lm_eval, wandb, and vllm; mergekit-evolve --help therefore
fails at import with a missing cma module. This is an explicit optional gate,
not a core mergekit import failure.
Read evolution.md only after the preflight. Never start a networked evaluation, W&B run, Ray cluster, vLLM server, or long CMA-ES search as a smoke test. Record seeds, max evaluations, output paths, evaluator, and stop conditions.
Run python scripts/check_extension_registration.py --help or the default
probe to inspect core registration without invoking optional evolution.