MLOps
模型部署、评估与运维
context-degradation
guanyang
This skill should be used when the user asks to "diagnose context problems", "fix lost-in-middle issues", "debug agent failures", "understand context poisoning", or mentions context degradation, attention patterns, context clash, context confusion, or agent performance degradation. Provides patterns for recognizing and mitigating context failures.
advanced-evaluation
guanyang
This skill should be used when the user asks to "implement LLM-as-judge", "compare model outputs", "create evaluation rubrics", "mitigate evaluation bias", or mentions direct scoring, pairwise comparison, position bias, evaluation pipelines, or automated quality assessment.
model-evaluation
cosmix
Evaluates machine learning models for performance, fairness, and reliability using appropriate metrics and validation techniques. Covers training debugging, hyperparameter tuning, and production monitoring. Trigger keywords: model evaluation, metrics, accuracy, precision, recall, F1, F1-score, ROC, AUC, ROC-AUC, confusion matrix, cross-validation, k-fold, stratified, overfitting, underfitting, bias, variance, bias-variance tradeoff, hyperparameter, hyperparameter tuning, loss, loss function, metric, benchmark, benchmarking, model performance, classification metrics, regression metrics, RMSE, MSE, MAE, MAPE, R2, R-squared, train-test split, validation set, test set, hold-out, learning curve, validation curve, model selection, error analysis, residual analysis, ML testing, training issues, convergence, gradient, vanishing gradient, exploding gradient, training instability, LLM evaluation, language model evaluation, prompt engineering evaluation, A/B testing, champion-challenger, model monitoring, model drift, data drift, concept drift, model decay.
debugging
cosmix
Systematically diagnoses and resolves software bugs, test failures, data quality issues, and performance problems using various debugging techniques and tools. Trigger keywords: debug, bug, error, exception, crash, issue, troubleshoot, fix, stack trace, diagnosis, diagnose, investigate, root cause, why, failing, broken, not working, unexpected, flaky, intermittent, regression, performance degradation.
multi-model-meta-analysis
petekp
Synthesize outputs from multiple AI models into a comprehensive, verified assessment. Use when: (1) User pastes feedback/analysis from multiple LLMs (Claude, GPT, Gemini, etc.) about code or a project, (2) User wants to consolidate model outputs into a single reliable document, (3) User needs conflicting model claims resolved against actual source code. This skill verifies model claims against the codebase, resolves contradictions with evidence, and produces a more reliable assessment than any single model.
r3f-loaders
EnzeD
React Three Fiber asset loading - useGLTF, useLoader, Suspense patterns, preloading. Use when loading 3D models, textures, HDR environments, or managing loading states.
BKUI 快速入门
TencentBlueKing
蓝鲸前端知识库入口指南，包含规范、索引和工作流程
fiftyone-model-evaluation
voxel51
Evaluate model predictions against ground truth using COCO, Open Images, or custom protocols. Use when computing mAP, precision, recall, confusion matrices, or analyzing TP/FP/FN examples for detection, classification, segmentation, or regression tasks.
fiftyone-embeddings-visualization
voxel51
Visualizes datasets in 2D using embeddings with UMAP or t-SNE dimensionality reduction. Use when exploring dataset structure, finding clusters, identifying outliers, or understanding data distribution.
llm-fine-tuning-guide
qodex-ai
Master fine-tuning of large language models for specific domains and tasks. Covers data preparation, training techniques, optimization strategies, and evaluation methods. Use when adapting models for specialized applications, reducing inference costs, or improving domain-specific performance.
baoyu-image-gen
Quriosity-agent
AI image generation with OpenAI, Google, DashScope, Replicate and fal.ai APIs. Supports text-to-image, reference images, aspect ratios. Sequential by default; parallel generation available on request. Use when user asks to generate, create, or draw images.
native-cli
Quriosity-agent
Run QCut's native TypeScript pipeline CLI for AI content generation, video analysis, transcription, YAML pipelines, ViMax agentic video production, and project management. Also use for editor HTTP automation tasks (state snapshots, events, transactions, and notification bridge control) when user needs deterministic state-aware control.
business-competitor-analysis
manojbajaj95
Perform comprehensive competitor analysis for any business. Produces an executive-summary markdown report with target customer profile, market positioning, pricing/business model, product features, funding/company size, SWOT analysis, and competitive matrix. All findings are data-grounded. Use when the user asks to analyze competitors, understand competitive landscape, compare a business to alternatives, or perform market research.
soul
saadshahd
Use when starting any task, thinking through confidence, verifying work, or asking "what could go wrong". Triggers on every non-trivial request, "how confident", "verify this", "think through".
arbos_cheap_model
unarbos
Run a cheap model (the Project's talking model) over text through the relay proxy, for map-reduce over big data.
israeli-pension-advisor
skills-il
Navigate the Israeli pension and savings system including pension funds (keren pensia), manager's insurance (bituach menahalim), training funds (keren hishtalmut), and retirement planning. Use when user asks about Israeli pension, "pensia", "keren hishtalmut", retirement savings, "bituach menahalim", pension contributions, or tax benefits from savings. Covers mandatory pension, voluntary savings, and withdrawal rules. Do NOT provide specific investment recommendations or fund performance comparisons.
litellm
BbgnsurfTech
When calling LLM APIs from Python code. When connecting to llamafile or local LLM servers. When switching between OpenAI/Anthropic/local providers. When implementing retry/fallback logic for LLM calls. When code imports litellm or uses completion() patterns.
aiconfig-update
launchdarkly
Update, archive, and delete LaunchDarkly AI Configs and their variations. Use when you need to modify config properties, change model parameters, update instructions or messages, archive unused configs, or permanently remove them.
trigger-agents
triggerdotdev
AI agent patterns with Trigger.dev - orchestration, parallelization, routing, evaluator-optimizer, and human-in-the-loop. Use when building LLM-powered tasks that need parallel workers, approval gates, tool calling, or multi-step agent workflows.
vscode-custom-agents
aktsmm
"VS Code カスタムエージェントの配置ルール・アクセス制御・マルチエージェント設計のベストプラクティス。Triggers on 'custom agent', 'agent not found', 'subagent', 'user-invokable', 'エージェント設計', 'ピッカー'."
slime
wundercorp
RL post-training for LLMs with Megatron and SGLang.
peft
wundercorp
Fine-tune large LLMs with LoRA on limited GPU memory.
trl-fine-tuning
wundercorp
"TRL: SFT, DPO, GRPO, RLOO reward modeling for LLM RLHF."
nemo-curator
wundercorp
"Curate LLM training data: dedupe, filter, PII redaction."