docker-model-runner
Skills for using Docker Model Runner to run local LLM inference
Skills for using Docker Model Runner to run local LLM inference
Domain knowledge for the Evolution Engine โ LLM-powered autonomous strategy discovery from raw OHLCV data. Covers the generate-backtest-select-evolve loop, vectorized backtesting, out-of-sample validation, and strategy graduation. Use when discovering trading patterns, running backtests, evolving strategies, or reviewing evolution logs. Triggers on "evolve", "discover patterns", "backtest", "evolution", "strategy generation", "candidate strategy".
A specialized skill for the sns-strategist and visual-planner agents covering platform algorithms. Provides how recommendation algorithms work and reach optimization strategies for Instagram, TikTok, LinkedIn, and X. Use for 'algorithm,' 'reach rate,' 'exposure optimization,' 'feed algorithm,' and similar topics.
Database normalization/denormalization pattern library. An extension skill for data-modeler that provides 1NF-BCNF criteria, functional dependency analysis, step-by-step normalization procedures, strategic denormalization patterns, and common domain ERD templates. Use when data modeling involves 'normalization', 'denormalization', 'ERD patterns', 'functional dependencies', 'table splitting', 'relationship design', etc. Note: DDL generation and query optimization are outside the scope of this skill.
Feature engineering techniques catalog: numeric/categorical/time-series/text transformations, feature selection, feature store design. Use this skill for data preprocessing and feature design involving 'feature engineering', 'variable transformation', 'encoding', 'scaling', 'feature selection', 'feature store', 'feature importance', etc. Enhances the data-engineer's feature engineering capabilities. Note: model design and training management are outside this skill's scope.
A full ML pipeline where an agent team collaborates to perform data preparation, model design, training, evaluation, and deployment readiness. Use this skill for 'design an ML experiment', 'train a model', 'machine learning project', 'build a deep learning model', 'classification model', 'regression model', 'data preprocessing', 'model evaluation', 'hyperparameter tuning', 'MLOps setup', 'XGBoost model', 'PyTorch model', and other ML experiment tasks. Supports data-preprocessing-only or evaluation-only requests as well. Note: model serving infrastructure (SageMaker/Vertex AI) direct deployment, large-scale distributed training cluster management, and real-time inference service operation are outside this skill's scope.
ML model selection matrix by problem type, hyperparameter tuning strategies, and ensemble methodology guide. Use this skill for ML model selection and design involving 'model selection', 'algorithm comparison', 'hyperparameter tuning', 'Optuna', 'ensemble', 'XGBoost vs LightGBM', 'model comparison', 'cross-validation', etc. Enhances the model-designer and evaluation-analyst's model design capabilities. Note: data preprocessing and training infrastructure management are outside this skill's scope.
Methodology for systematically evaluating and optimizing LLM prompt quality. Use this skill for 'prompt optimization', 'prompt improvement', 'guardrail design', 'prompt debugging', 'few-shot optimization', 'system prompt design', and other prompt quality improvement tasks. Note: LLM model fine-tuning and model weight modification are outside the scope of this skill.
A specialized skill providing CEFR (Common European Framework of Reference) language proficiency assessment matrices and diagnostic tools. Used by the level-assessor agent for precise per-skill level diagnosis and learning gap analysis. Automatically applied in contexts such as 'CEFR level', 'level test', 'language proficiency assessment', 'A1-C2', 'per-skill diagnosis'. However, official exam certification issuance and formal CEFR certification are outside the scope of this skill.
A specialized skill for systematically classifying error patterns and constructing concept deficit maps. Used by error-analyst and diagnostician agents when analyzing learner error data to diagnose root causes and develop customized remediation strategies. Automatically applied in contexts such as 'error patterns', 'wrong answer analysis', 'concept deficits', 'weakness analysis', 'error types'. However, psychological testing and learning disability diagnosis are outside the scope of this skill.
A specialized skill for building KSA (Knowledge, Skill, Ability) taxonomies in job analysis and mapping them to NCS/O*NET standards. Used by the job-analyst and competency-architect agents when writing job descriptions and systematically classifying competencies. Automatically applied in contexts involving 'KSA analysis,' 'job description,' 'NCS,' 'O*NET,' 'task analysis,' or 'competency classification system.' Note: actual NCS system access or HR system data integration is outside the scope of this skill.
MQM(Multidimensional Quality Metrics) ๊ธฐ๋ฐ ๋ฒ์ญ ํ์ง ํ๊ฐ ํ๋ ์์ํฌ, ์ค๋ฅ ๋ถ๋ฅ ์ฒด๊ณ, ์ฌ๊ฐ๋ ๊ฐ์ค์น, ์ ์ ์ฐ์ถ ๊ณต์์ ์ ๊ณตํ๋ quality-reviewer ํ์ฅ ์คํฌ. 'MQM ํ๊ฐ', '๋ฒ์ญ ํ์ง ์ ์', '์ค๋ฅ ๋ถ๋ฅ', '๋ฒ์ญ QA', 'ํ์ง ๋ฉํธ๋ฆญ' ๋ฑ ๋ฒ์ญ ํ์ง์ ์ ๋์ ๊ฒ์ฆ์ด ํ์ํ ๋ ์ฌ์ฉํ๋ค. ๋จ, ๋ฒ์ญ ์์ฒด ์ํ์ด๋ ํ์งํ ์ ์ฉ์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
๋ฐ์ดํฐ๋ฒ ์ด์ค ์ ๊ทํ/๋น์ ๊ทํ ํจํด ๋ผ์ด๋ธ๋ฌ๋ฆฌ. 1NF~BCNF ํ๋ณ ๊ธฐ์ค, ํจ์ ์ข ์์ฑ ๋ถ์, ์ ๊ทํ ๋จ๊ณ๋ณ ๋ณํ ์ ์ฐจ, ์ ๋ต์ ๋น์ ๊ทํ ํจํด, ๊ณตํต ๋๋ฉ์ธ ERD ํ ํ๋ฆฟ์ ์ ๊ณตํ๋ data-modeler ํ์ฅ ์คํฌ. '์ ๊ทํ', '๋น์ ๊ทํ', 'ERD ํจํด', 'ํจ์ ์ข ์์ฑ', 'ํ ์ด๋ธ ๋ถ๋ฆฌ', '๊ด๊ณ ์ค๊ณ' ๋ฑ ๋ฐ์ดํฐ ๋ชจ๋ธ๋ง ์ ์ฌ์ฉํ๋ค. ๋จ, DDL ์์ฑ์ด๋ ์ฟผ๋ฆฌ ์ต์ ํ๋ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
ML ์คํ์ ๋ฐ์ดํฐ ์ค๋น, ๋ชจ๋ธ ์ค๊ณ, ํ์ต, ํ๊ฐ, ๋ฐฐํฌ ์ค๋น๋ฅผ ์์ด์ ํธ ํ์ด ํ์ ํ์ฌ ์ํํ๋ ํ ML ํ์ดํ๋ผ์ธ. 'ML ์คํ ์ค๊ณํด์ค', '๋ชจ๋ธ ํ์ตํด์ค', '๋จธ์ ๋ฌ๋ ํ๋ก์ ํธ', '๋ฅ๋ฌ๋ ๋ชจ๋ธ ๋ง๋ค์ด์ค', '๋ถ๋ฅ ๋ชจ๋ธ', 'ํ๊ท ๋ชจ๋ธ', '๋ฐ์ดํฐ ์ ์ฒ๋ฆฌ', '๋ชจ๋ธ ํ๊ฐ', 'ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋', 'MLOps ์ค์ ', 'XGBoost ๋ชจ๋ธ', 'PyTorch ๋ชจ๋ธ' ๋ฑ ML ์คํ ์ ๋ฐ์ ์ด ์คํฌ์ ์ฌ์ฉํ๋ค. ๋ฐ์ดํฐ ์ ์ฒ๋ฆฌ๋ง ํ์ํ๊ฑฐ๋ ๋ชจ๋ธ ํ๊ฐ๋ง ํ์ํ ๊ฒฝ์ฐ์๋ ์ง์ํ๋ค. ๋จ, ๋ชจ๋ธ ์๋น ์ธํ๋ผ(SageMaker/Vertex AI) ์ง์ ๋ฐฐํฌ, ๋๊ท๋ชจ ๋ถ์ฐ ํ์ต ํด๋ฌ์คํฐ ๊ด๋ฆฌ, ์ค์๊ฐ ์ถ๋ก ์๋น์ค ์ด์์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
ML ๋ฌธ์ ์ ํ๋ณ ๋ชจ๋ธ ์ ํ ๋งคํธ๋ฆญ์ค, ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋ ์ ๋ต, ์์๋ธ ๋ฐฉ๋ฒ๋ก ๊ฐ์ด๋. '๋ชจ๋ธ ์ ํ', '์๊ณ ๋ฆฌ์ฆ ๋น๊ต', 'ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋', 'Optuna', '์์๋ธ', 'XGBoost vs LightGBM', '๋ชจ๋ธ ๋น๊ต', '๊ต์ฐจ ๊ฒ์ฆ' ๋ฑ ML ๋ชจ๋ธ ์ ํ ๋ฐ ์ค๊ณ ์ ์ด ์คํฌ์ ์ฌ์ฉํ๋ค. model-designer์ evaluation-analyst์ ๋ชจ๋ธ ์ค๊ณ ์ญ๋์ ๊ฐํํ๋ค. ๋จ, ๋ฐ์ดํฐ ์ ์ฒ๋ฆฌ๋ ํ์ต ์ธํ๋ผ ๊ด๋ฆฌ๋ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
CEFR(์ ๋ฝ๊ณตํต์ฐธ์กฐ๊ธฐ์ค) ๊ธฐ๋ฐ ์ธ์ด ๋ฅ๋ ฅ ํ๊ฐ ๋งคํธ๋ฆญ์ค์ ์ง๋จ ๋๊ตฌ๋ฅผ ์ ๊ณตํ๋ ์ ๋ฌธ ์คํฌ. level-assessor ์์ด์ ํธ๊ฐ ํ์ต์์ ์์ญ๋ณ ๋ ๋ฒจ์ ์ ๋ฐ ์ง๋จํ๊ณ ํ์ต ๊ฐญ์ ๋ถ์ํ ๋ ํ์ฉํ๋ค. 'CEFR ๋ ๋ฒจ', '๋ ๋ฒจ ํ ์คํธ', '์ธ์ด ๋ฅ๋ ฅ ํ๊ฐ', 'A1-C2', '์์ญ๋ณ ์ง๋จ' ๋ฑ์ ๋งฅ๋ฝ์์ ์๋ ์ ์ฉํ๋ค. ๋จ, ๊ณต์ธ ์ํ ์ธ์ฆ์ ๋ฐ๊ธ์ด๋ ๊ณต์ CEFR ์ธ์ฆ์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
์ธ๊ตญ์ด ํ์ต ํ ํ์ดํ๋ผ์ธ. ๋ ๋ฒจ ํ ์คํธโ์ปค๋ฆฌํ๋ผ ์ค๊ณโ๋ ์จ ์์ฑโํด์ฆ ์ถ์ โ๋ณต์ต ๊ด๋ฆฌ๋ฅผ ์์ด์ ํธ ํ์ด ํ์ ํ์ฌ ์ ๊ณตํ๋ค. '์์ด ๊ณต๋ถ', '์ผ๋ณธ์ด ํ์ต', '์ธ๊ตญ์ด ๋ฐฐ์ฐ๊ธฐ', '์ธ์ด ํ์ต', '์์ด ๋ฌธ๋ฒ', 'ํํ ์ฐ์ต', '๋ ๋ฒจ ํ ์คํธ', '์ดํ ์๊ธฐ', '์์ด ๋ ํ', '์ค๊ตญ์ด ์ ๋ฌธ', 'TOEIC ์ค๋น', '์์ด ํํ' ๋ฑ ์ธ๊ตญ์ด ํ์ต ์ ๋ฐ์ ์ด ์คํฌ์ ์ฌ์ฉํ๋ค. ํน์ ์์ญ๋ง ํ์ตํ๋ ๊ฒ๋ ์ง์ํ๋ค. ๋จ, ์ค์๊ฐ ์์ฑ ๋ํ(STT/TTS), ์์ด๋ฏผ ๋งค์นญ, ๊ณต์ธ ์ํ ์์ ๋ํ์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
ํ ๋ก ๋ ผ์ฆ ๊ตฌ์ถ์ ํ์ํ Toulmin ๋ ผ์ฆ ๋ชจ๋ธ, ์ฆ๊ฑฐ ์ ํ ๋ถ๋ฅ, ๋ฐ๋ฐ ์ ๋ต์ ์ฒด๊ณ์ ์ผ๋ก ์ ๊ณตํ๋ ์ ๋ฌธ ์คํฌ. pro-debater์ con-debater๊ฐ ๊ฒฌ๊ณ ํ ๋ ผ์ฆ์ ๊ตฌ์ถํ๊ณ ์๋ ๋ ผ๊ฑฐ๋ฅผ ํจ๊ณผ์ ์ผ๋ก ๋ฐ๋ฐํ ๋ ํ์ฉํ๋ค. '๋ ผ์ฆ ๊ตฌ์กฐ', 'Toulmin ๋ชจ๋ธ', '๋ ผ๊ฑฐ ๊ตฌ์ถ', '๋ฐ๋ฐ ์ ๋ต', '์ฆ๊ฑฐ ์ ํ', '๋ ผ์ฆ ๊ฐํ' ๋ฑ์ ๋งฅ๋ฝ์์ ์๋ ์ ์ฉํ๋ค. ๋จ, ๋ฒ๋ฅ ์ ๋ณ๋ก ์ด๋ ํ์ ๋ ผ๋ฌธ์ ๋ ผ์ฆ ๊ตฌ์กฐ ์์ฑ์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
์ง๋ฌด ๋ถ์์์ KSA(Knowledge, Skill, Ability) ์ฒด๊ณ๋ฅผ ๊ตฌ์ถํ๊ณ NCS/O*NET ํ์ค๊ณผ ๋งคํํ๋ ์ ๋ฌธ ์คํฌ. job-analyst์ competency-architect๊ฐ ์ง๋ฌด๊ธฐ์ ์๋ฅผ ์์ฑํ๊ณ ์ญ๋์ ์ฒด๊ณ์ ์ผ๋ก ๋ถ๋ฅํ ๋ ํ์ฉํ๋ค. 'KSA ๋ถ์', '์ง๋ฌด๊ธฐ์ ์', 'NCS', 'O*NET', '๊ณผ์ ๋ถ์', '์ญ๋ ๋ถ๋ฅ ์ฒด๊ณ' ๋ฑ์ ๋งฅ๋ฝ์์ ์๋ ์ ์ฉํ๋ค. ๋จ, ์ค์ NCS ์์คํ ์ ์์ด๋ ์ธ์ฌ ์์คํ ๋ฐ์ดํฐ ์ฐ๋์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
์๋๊ณต์ ยท์ธ์ก๊ณต์ ํญ๋ชฉ์ ์ต์ ์กฐํฉ์ผ๋ก ๊ทน๋ํํ๋ ๊ณต์ ์ต์ ํ ์์ง. 'deduction-optimizer' ์์ด์ ํธ๊ฐ ๊ณต์ ํญ๋ชฉ์ ๋ถ์ํ๊ณ ์ต์ ์กฐํฉ์ ๋์ถํ ๋ ์ด ์คํฌ์ ๊ณต์ ํ๋ํ, ์ฐ์ ์์ ์๊ณ ๋ฆฌ์ฆ, ์๋ฎฌ๋ ์ด์ ๋ฐฉ๋ฒ์ ๋ฐ๋์ ํ์ฉํด์ผ ํ๋ค. '๊ณต์ ์ต์ ํ', '์ธ์ก๊ณต์ ๊ทน๋ํ', '์ฐ๋ง์ ์ฐ ์๋ฎฌ๋ ์ด์ ' ๋ฑ์ ์ฌ์ฉํ๋ค. ๋จ, ์๋ ๋ถ๋ฅ๋ ์ธ์ก ๊ณ์ฐ์ ์ด ์คํฌ์ ๋ฒ์๊ฐ ์๋๋ค.
์ญ๋ ๋ชจ๋ธ ์ค๊ณ ํ๋ ์์ํฌ. jd-writer์ screening-expert ์์ด์ ํธ๊ฐ ์ง๋ฌด๋ณ ํต์ฌ ์ญ๋์ ์ ์ํ๊ณ ํ๊ฐ ๊ธฐ์ค์ ์๋ฆฝํ ๋ ์ฐธ์กฐ. '์ญ๋ ๋ชจ๋ธ', '์ง๋ฌด ์ญ๋', '์คํฌ ๋งคํธ๋ฆญ์ค' ์์ฒญ ์ ์ฌ์ฉ. ๋จ, ์ธ์ฌ ํ๊ฐ ์์คํ ๊ตฌ์ถ์ด๋ ๊ธ์ฌ ์ฒด๊ณ ์ค๊ณ๋ ๋ฒ์ ๋ฐ.
Selects a base model and fine-tuning technique (SFT, DPO, or RLVR) for the user's use case by querying SageMaker Hub. Use when the user asks which model or technique to use, wants to start fine-tuning, or mentions a model name or family (e.g., "Llama", "Mistral") โ always activate even for known model names because the exact Hub model ID must be resolved. Queries available models, validates technique compatibility, and confirms selections.