"Extend mergekit with custom merge methods and task graphs, or plan
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill extension-and-evolution --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Extension And Evolution?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-extension-and-evolution)More formats (shields.io, HTML) on the badges page.
---
name: extension-and-evolution
description: "Extend mergekit with custom merge methods and task graphs, or plan
optional evolutionary merge searches with explicit dependency and evaluation
gates."
disable-model-invocation: true
metadata:
disco-role: operating
license: LGPL 3.0
---
# Extension and evolution route
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.
## Route first
- Use [merge-configs](../merge-configs/SKILL.md) for ordinary merge YAML,
registered method selection, parameters, tokenizer, and chat templates.
- Use [model-io-and-architecture](../model-io-and-architecture/SKILL.md) for
model references, architecture conversion, tensor IO, devices, and memory.
- Use [specialized-workflows](../specialized-workflows/SKILL.md) for raw,
MoE, multi-stage, LoRA, tokensurgeon, or layer-shuffle commands.
## Custom method procedure
1. Define the mathematical contract and required parameters before choosing the
decorator/easy-define or class-based implementation.
2. Implement a tensor-local method that respects mergekit's task and device
contract; register it under the exact YAML method name.
3. Check registration in a clean import, exercise a tiny tensor fixture, and
add focused tests for defaults, parameter validation, dtype, and device.
4. Run the graph tests and a focused package test before using the method in a
real model merge. Keep the source method self-contained and documented.
Read [custom-methods.md](references/custom-methods.md) for the API patterns and
[troubleshooting.md](references/troubleshooting.md) for signature, registry,
and task-graph failures.
## Evolution procedure
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](references/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.
## Safe helper
Run `python scripts/check_extension_registration.py --help` or the default
probe to inspect core registration without invoking optional evolution.

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!