model-merging

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Quality

76%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/model-merging/SKILL.md

The canonical home for this skill is model-merging in OpenLAIR/dr-claw

Quality

Content

65%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

Highly actionable content with comprehensive executable configs, but it is verbose due to duplicated method examples and lacks an explicit validated merge workflow. Moving duplicated method detail into the existing reference files would improve both conciseness and progressive disclosure.

Suggestions

Consolidate the duplicated linear/SLERP configs: keep one minimal example in Quick Start and defer the full method catalog to references/methods.md instead of repeating it in Merge Methods Guide.

Add an explicit numbered merge workflow with a validation checkpoint (e.g., write config -> run mergekit-yaml -> benchmark on MT-Bench/MMLU -> only deploy if scores hold) rather than burying the 'always benchmark' guidance in Common Pitfalls.

Trim concept primers (what SLERP, DARE, and Task Arithmetic are) to one-line definitions and rely on the reference files for depth, since Claude already knows these algorithms.

Dimension	Reasoning	Score
Conciseness	Mostly efficient but visibly padded by duplication: linear and SLERP configs appear in both Quick Start and Merge Methods Guide, method explanations repeat across sections, and several concept primers (what SLERP/DARE/Task Arithmetic are) could be trimmed since Claude already knows them.	3 / 5
Actionability	Fully executable, copy-paste ready mergekit YAML configs and bash/python commands covering all common cases (linear, SLERP, task arithmetic, TIES, DARE, MoE, deployment, quantization).	5 / 5
Workflow Clarity	A rough install-config-merge-evaluate-deploy sequence is implied across sections, but there is no explicit numbered workflow with validation checkpoints; merging is a batch operation, and the 'always benchmark before deploying' guidance lives in Common Pitfalls rather than as an explicit validate-fix-retry loop, capping this at 3.	3 / 5
Progressive Disclosure	Three real reference files (methods.md, examples.md, evaluation.md) are signaled in a See Also section, but the body inlines ~540 lines of method configs, advanced patterns, and pitfalls that overlap with those references, so content that should be separate remains inline.	3 / 5
	Total	14 / 20 Passed

Description

87%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, well-structured description that clearly states what the skill does and when to use it with concrete trigger phrases. Minor room for improvement in replacing motivational phrasing with concrete actions and adding a few more synonyms.

Dimension	Reasoning	Score
Specificity	Lists several concrete actions ('Merge multiple fine-tuned models using mergekit', 'combine capabilities without retraining', 'blending domain-specific expertise', 'production deployment strategies'), but phrases like 'improving performance beyond single models' and 'experimenting rapidly with model variants' are motivations rather than concrete actions, leaving minor gaps.	4 / 5
Completeness	Explicitly answers both what ('Merge multiple fine-tuned models using mergekit to combine capabilities without retraining') and when ('Use when creating specialized models by blending domain-specific expertise... or experimenting rapidly with model variants') with concrete trigger phrases.	5 / 5
Trigger Term Quality	Good natural keyword coverage ('merge multiple fine-tuned models', 'specialized models', 'blending domain-specific expertise', 'math + coding + chat', 'model variants') that users would plausibly say, though a few common synonyms like 'combine models' or 'fuse models' are absent.	4 / 5
Distinctiveness Conflict Risk	Occupies a clear niche (mergekit-based model merging) with distinct triggers and tool-specific framing, making conflict with other skills unlikely.	5 / 5
	Total	18 / 20 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (541 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/model-merging/SKILL.md
Commit: 3a6c3a9

Reviewed: 1 day ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.