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.
Generate images with Pruna P-Image models via inference.sh CLI. Models: P-Image, P-Image-LoRA, P-Image-Edit, P-Image-Edit-LoRA. Capabilities: text-to-image, image editing, LoRA styles, multi-image compositing, fast inference. Pruna optimizes models for speed without quality loss. Triggers: pruna, p-image, pruna image, fast image generation, optimized flux, pruna ai, p image, fast ai image, economic image generation, cheap image generation
Generate and edit images with Alibaba Qwen-Image-2.0 models via inference.sh CLI. Models: Qwen-Image-2.0 (fast), Qwen-Image-2.0-Pro (professional text rendering). Capabilities: text-to-image, multi-image editing, complex text rendering. Triggers: qwen image, qwen-image, alibaba image, dashscope image, qwen image 2, qwen image pro
Implement ReasoningBank adaptive learning with AgentDB's 150x faster vector database. Includes trajectory tracking, verdict judgment, memory distillation, and pattern recognition. Use when building self-learning agents, optimizing decision-making, or implementing experience replay systems.
A prompt repetition technique for improving LLM accuracy. Achieves significant performance gains in 67% (47/70) of 70 benchmarks. Automatically applied on lightweight models (haiku, flash, mini).
Generate and edit images with Alibaba Qwen-Image-2.0 models via inference.sh CLI. Models: Qwen-Image-2.0 (fast), Qwen-Image-2.0-Pro (professional text rendering). Capabilities: text-to-image, multi-image editing, complex text rendering. Triggers: qwen image, qwen-image, alibaba image, dashscope image, qwen image 2, qwen image pro
Data validation and pipeline testing utilities for ML training projects. Validates datasets, model checkpoints, training pipelines, and dependencies. Use when validating training data, checking model outputs, testing ML pipelines, verifying dependencies, debugging training failures, or ensuring data quality before training.
Build and train machine learning models using scikit-learn, PyTorch, and TensorFlow for classification, regression, and clustering tasks
Extract an Allium specification from an existing codebase. Use when the user has existing code and wants to distil behaviour into a spec, reverse engineer a specification from implementation, generate a spec from code, turn implementation into a behavioural specification, or document what a codebase does in Allium terms.
Enables a secondary AI model to advise the primary when it gets stuck, fails repeatedly, or needs upfront planning.
Create and train AI learning plugins with AgentDB's 9 reinforcement learning algorithms. Includes Decision Transformer, Q-Learning, SARSA, Actor-Critic, and more. Use when building self-learning agents, implementing RL, or optimizing agent behavior through experience.
Implement GRPO (Group Relative Policy Optimization) fine-tuning for vision-language models on small datasets. Use when SFT underperforms or training data is limited (<1000 examples).
Agent-Native PMF 产品市场匹配验证领域，用多 Agent 模拟替代高成本真实流量测试，预判产品与市场的匹配程度。
Sibyl pilot 验证决策 agent - 根据小实验结果决定 ADVANCE / REFINE / PIVOT
Sibyl 监督决策 agent - 分析实验结果决定 PIVOT 还是 PROCEED
This skill should be used when users want to fine-tune language models or perform reinforcement learning (SFT, DPO, GRPO, ORPO, KTO, SimPO) using the highly optimized Unsloth library. Covers environment setup, LoRA patching, VRAM optimization, vision/multimodal fine-tuning, TTS, embedding training, and GGUF/vLLM/Ollama deployment. Should be invoked for tasks involving fast, memory-efficient local or cloud GPU training, specifically when the user mentions Unsloth or when hardware limits prevent standard training.
Evidence-first retrieval-augmented reasoning skill for decision-critical scenarios. Retrieves information from a specified knowledge base, extracts verifiable evidence, detects conflicting claims, and evaluates answer sufficiency with explicit confidence and risk signals. Produces traceable outputs suitable for agent-level decision control and escalation.
Trigger Pattern Vault/pool/first-depositor pattern detected - Inject Into Depth-edge-case
Performs comprehensive token flow analysis by tracing all token entry and exit paths, verifying accounting consistency, detecting unsolicited transfer vectors, and identifying risks such as donation attacks, balance desynchronization, token type confusion, and side-effect-driven state changes.
Convert a backtesting notebook into a parameter optimisation notebook using the bundled transformation mapping to choose searchable parameters and rewrite the notebook.
Run multiple AI models in parallel for 3-5x speedup with ENFORCED performance statistics tracking. Use when validating with Grok, Gemini, GPT-5, DeepSeek, MiniMax, Kimi, GLM, or Claudish proxy for code review, consensus analysis, or multi-expert validation. NEW in v3.2.0 - Direct API prefixes (mmax/, kimi/, glm/) for cost savings. Includes dynamic model discovery via `claudish --top-models` and `claudish --free`, session-based workspaces, and Pattern 7-8 for tracking model performance. Trigger keywords - "grok", "gemini", "gpt-5", "deepseek", "minimax", "kimi", "glm", "claudish", "multiple models", "parallel review", "external AI", "consensus", "multi-model", "model performance", "statistics", "free models".
