What are the best Claude Code skills for LLM evals?
advanced-evaluation from muratcankoylan/Agent-Skills-for-Context-Engineering is the highest-scoring Claude Code skill for LLM evals, at 86/100 (grade B) on the Suede Skill Check. The table below compares the top 15 of 17 listed LLM eval skills from 11 repositories by score, publisher and license, using the October 22, 2026 snapshot of SuedeAI.directory.
- 17 listed LLM eval skills
- 5 full pages scoring 80+
- 12 repositories
Compare 15 Claude Code skills for LLM evals
Ranked by Suede Skill Check score. Linked names open the directory listing; scores come from the published rubric.
| Skill | Publisher | Suede Skill Check | License | What it does |
|---|---|---|---|---|
| advanced-evaluation | muratcankoylan muratcankoylan/Agent-Skills-for-Context-Engineering | 86/100 · B | MIT | Guides advanced LLM evaluation, including LLM-as-judge systems, direct scoring, pairwise comparison, rubric calibration, evaluator bias mitigation, confidence scoring, and automated quality assessment. |
| evaluation | muratcankoylan muratcankoylan/Agent-Skills-for-Context-Engineering | 86/100 · B | MIT | Guides building agent evaluation systems with deterministic checks, regression suites, multi-dimensional rubrics, quality gates, production monitoring, baseline comparison, and outcome measurement for agent pipelines. |
| huggingface-community-evals | Hugging Face Official huggingface/skills | 86/100 · B | Apache-2.0 | Runs evaluations for Hugging Face Hub models on local hardware using inspect-ai and lighteval, including backend selection and choosing between vLLM, Transformers, and accelerate. |
| prompt-engineer | jeffallan jeffallan/claude-skills | 84/100 · B | MIT | Writes, refactors, and evaluates prompts for LLMs — generating optimized prompt templates, structured output schemas, evaluation rubrics, and test… |
| skill-creator | Anthropic Official anthropics/skills | 83/100 · B | Apache-2.0 | Guides creating new skills, improving existing ones, running evals, benchmarking skill performance with variance analysis, and optimizing a skill's description for better triggering. |
| eval-driven-dev | GitHub Official github/awesome-copilot | 82/100 · B | MIT | Improve AI application with evaluation-driven development. Define eval criteria, instrument the application, build golden datasets, observe and… |
| eval-engineering | langchain-ai Official langchain-ai/langchain-skills | 82/100 · B | MIT | Inspect an agent repository and optional traces, interview the user, write reviewed Task Specs, build and audit Harbor tasks, and bootstrap reusable… |
| rag-eval | NVIDIA Official NVIDIA/skills | 81/100 · B | Apache-2.0 | Filesystem RAG benchmarks: corpus/, train.json, evaluate_rag.py (RAGAS quality). Not for prod monitoring, latency/throughput benchmarking (use… |
| prompt-optimizer | Sentry Official getsentry/skills | 81/100 · B | Apache-2.0 | Creates and refines agent, system, and developer prompts and reusable prompt templates, including porting prompts between OpenAI, Claude, and Gemini and building prompt evals. |
| digital-health-clinical-asr-eval | NVIDIA Official NVIDIA/skills | 79/100 · C | Apache-2.0 | Stage 3 of Clinical ASR Flywheel. Score a NeMo manifest, produce the five-section KER leaderboard (by-ipa_source diagnostic). Not for ASR auth… |
| agentic-eval | GitHub Official github/awesome-copilot | 77/100 · C | MIT | Patterns and techniques for evaluating and improving AI agent outputs. Use this skill when: - Implementing self-critique and reflection loops… |
| agent-platform-eval-flywheel | Google Official google/skills | 74/100 · C | Apache-2.0 | Measures and improves the quality of AI models and agents on Google Cloud using the Eval Quality Flywheel methodology. Use when generating synthetic… |
| decision-eval | avelikiy avelikiy/great_cto | 71/100 · C | MIT | Spawns the decision-scorer agent after architect proposes 2+ variants in an ADR. Produces a weighted scoring table and recommended choice saved to… |
| evaluating-llms-harness | Orchestra-Research Orchestra-Research/AI-Research-SKILLs | 69/100 · D | MIT | Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing… |
| nemo-evaluator-sdk | Orchestra-Research Orchestra-Research/AI-Research-SKILLs | 69/100 · D | MIT | Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing… |
Questions and answers
Which LLM eval skill has the highest Suede Skill Check score?
advanced-evaluation from muratcankoylan/Agent-Skills-for-Context-Engineering scores 86/100, grade B, the highest of the 17 listed LLM eval skills in the October 22, 2026 snapshot. It earns full points for structural checks, recent maintenance and instruction depth. evaluation from muratcankoylan/Agent-Skills-for-Context-Engineering follows at 86/100.
How do I install an LLM eval skill in Claude Code?
Run npx skills add with the skill folder's GitHub URL from your project, or copy the folder into .claude/skills/ and Claude Code loads it when a request matches its description. For advanced-evaluation, run: npx skills add https://github.com/muratcankoylan/Agent-Skills-for-Context-Engineering/tree/main/skills/advanced-evaluation. Each full skill page on SuedeAI.directory lists the same steps for Claude Code and Codex.
Are these LLM eval skills free?
All 15 LLM eval skills on this page are published in public GitHub repositories and cost nothing to download. Every one uses an open-source or open-content license, led by MIT (9) and Apache-2.0 (6), so you can use and adapt the files at no cost under those terms. One skill names credentials for an outside service; each full page lists them under Before you install.
Do these LLM eval skills work with Codex and Cursor?
Yes. Each of the 15 LLM eval skills on this page is a SKILL.md folder in the Agent Skills format, which Claude Code, Codex and Cursor all read. The skills CLI (npx skills add) installs every one of them into Claude Code, Codex, Cursor and other agents, and each full skill page adds a project-level copy into .claude/skills for Claude Code and .agents/skills for Codex. 3 of them declare compatibility in the SKILL.md frontmatter.
How does SuedeAI.directory rank LLM eval skills?
By Suede Skill Check score from 0 to 99, which adds structure, description quality, instruction depth, bundled files, maintenance and curated-list citations, all read from each skill's published files. A skill joins this page when its name or summary matches the LLM evals keyword map; each repository fills at most three rows, so the table compares publishers.
Which LLM eval skills come from official publishers?
huggingface-community-evals from Hugging Face, skill-creator from Anthropic, eval-driven-dev and agentic-eval from GitHub, eval-engineering from langchain-ai, rag-eval and digital-health-clinical-asr-eval from NVIDIA, prompt-optimizer from Sentry and agent-platform-eval-flywheel from Google are published by the company behind the tool, marked Official in the table.
Compare alternatives
More Claude Code skills by task
Presentations · Playwright and Browser Testing · Terraform · SEO · Code Review · Git and Pull Requests · MCP Servers · Data Analysis · Kubernetes · AWS · Security Audits · Email · Video · iOS and Swift · Flutter and Dart · React and Next.js · RAG and Vector Search · SQL and Databases · Technical Documentation · UI Design
Browse by category: AI Agents and LLM Engineering.