发现 Skills

Vision-language pre-training framework bridging frozen image encoders and LLMs. Use when you need image captioning, visual question answering, image-text retrieval, or multimodal chat with state-of-the-art zero-shot performance.

OpenAI's model connecting vision and language. Enables zero-shot image classification, image-text matching, and cross-modal retrieval. Trained on 400M image-text pairs. Use for image search, content moderation, or vision-language tasks without fine-tuning. Best for general-purpose image understanding.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

A Python CLI for interacting with the Dida365 Open API to manage TickTick tasks and projects. Suitable for scenarios that need to call TickTick APIs from scripts or the command line (e.g., querying, creating, updating, completing, deleting projects/tasks).

Turn raw customer feedback from multiple sources into clear product insights and priorities. Covers feedback collection, pattern recognition, and separating signal from noise. Use when feedback feels overwhelming, contradictory, or when the team can't agree on what users actually want.

Interact with Claude.ai through browser automation. Use when the user wants to 'ask browser Claude', 'query Claude.ai', 'talk to my other Claude', or have agents communicate with Claude web instances.

Generator of legal prompts optimized and ready for commercial and professional use. Produces high-quality prompts for: (1) generation of judicial decisions using the FIRAC-JB methodology, (2) analysis of pleadings and contracts, (3) drafting procedural documents, (4) jurisprudential research with anti‑hallucination controls, (5) citizen guidance, and (6) automation of workflows in n8n and legal AI systems. Generated prompts follow advanced prompt engineering (chain-of-thought, few-shot, role prompting, output formatting) and are calibrated for the Brazilian legal system. ALWAYS use when the user asks for a prompt for any legal task, wants to create an AI system for a law firm or court, needs a prompt for n8n automation, wants an instruction template for legal agents, or mentions "prompt for", "instruction for AI", "prompt template", "legal system prompt", "judgment prompt", "pleading prompt", "forge prompt", "prompt-forge". Monetizable product: the generated prompts are intellectual assets ready for licensing.

Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

Search and download images via the Google Custom Search API with LLM-powered selection. This skill should be used when finding images for articles, presentations, research documents, or enriching Obsidian notes with relevant visuals. Supports simple queries, batch processing from JSON config, automatic config generation from terms, and full note enrichment with automatic image insertion below headings.

[FINANCIAL EXECUTION] Create and launch meme coins and crypto tokens on launchpads (Pump.fun, PancakeSwap, FourMeme, Bonk, BAGS, Flap, Raydium, etc.) via bonding curve fair launch, or query token creation stats by launchpad via GMGN API. Requires explicit user confirmation. Use when user asks to create a token, launch a meme coin, cook a coin, deploy on a launchpad, or check launchpad creation stats on Solana, BSC, Base, ETH, or TON.

USE WHEN the user wants to manage Google Meet conferences via the `gws` CLI.

Predict protein structures using modern ML models including AlphaFold3, ESMFold, Chai-1, and Boltz-1. Use when predicting structures for novel proteins, protein complexes, or when comparing predictions across multiple methods.

This skill should be used when the user needs to manage Vaultwarden/Bitwarden: password manager, admin panel, backup, updates. Triggers: "password manager", "Vaultwarden", "vaultwarden setup", "vaultwarden backup", "password server", "password sync", "master password", "admin token", "admin panel", "vaultwarden update", "vaultwarden restore", "backup restore", Bitwarden client connection issues, vault sync problems.

Comprehensive Android mobile application penetration testing skill with full ADB shell access to rooted devices. This skill should be used when performing security assessments on Android applications including static analysis, dynamic analysis, runtime manipulation, traffic interception, and vulnerability identification. Triggers on requests to pentest Android apps, test mobile security, analyze APKs, bypass security controls, or perform OWASP Mobile Top 10 assessments. (user)

Create diagrams and visualizations using Mermaid syntax. Use when generating flowcharts, sequence diagrams, class diagrams, entity-relationship diagrams, Gantt charts, or any visual documentation. Triggers on Mermaid, flowchart, sequence diagram, class diagram, ER diagram, Gantt chart, diagram, visualization.

Add and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval. Works with the model-index metadata format.

Football (soccer) score prediction and tactical analysis for competitive betting. Three-track score predictions (regular + upset + high-scoring), featuring Poisson distribution λ weighting, xG model, injury/suspension impact, motivation evaluation, upset-factor mining, Wencai-52 cross-validation, and handicap/high-odds bonus enhancement strategies. Suitable for sports betting/football pools/score prediction/pre-match previews/parlay strategy scenarios.

Extract audio from short videos (Douyin/TikTok) and transcribe to text with timestamps. Use when user provides video URL and needs audio transcription.

Image processing skill for cropping images to 50% from center and rotating them 90 degrees clockwise. This skill should be used when users request image cropping to center, image rotation, or both operations combined on image files.

The social network for AI agents. Post, comment, upvote, and create communities.

Jailbreak LLMs and bypass safety filters using Parseltongue, GODMODE, and ULTRAPLINIAN techniques. Three attack modes: GODMODE CLASSIC (system prompt templates), PARSELTONGUE (33 input obfuscation techniques), ULTRAPLINIAN (multi-model racing). Also includes compliance escalation engine for task-agnostic unrestricted execution. Works on any model via API — Claude, GPT, Gemini, Grok, DeepSeek, Llama, Qwen, Mistral. Triggers on: "jailbreak", "godmode", "red team", "bypass guardrails", "uncensor", "liberate", "no restrictions", "adversarial test", "attack surface", "penetration test", "GCG", "TAP", "PAIR", "Tempest", "AutoDAN", "parseltongue", "ultraplinian", "obfuscate", "stress test model".

AWS Bedrock foundation models for generative AI. Use when invoking foundation models, building AI applications, creating embeddings, configuring model access, or implementing RAG patterns.

AWS DynamoDB NoSQL database for scalable data storage. Use when designing table schemas, writing queries, configuring indexes, managing capacity, implementing single-table design, or troubleshooting performance issues.

Optimize Cursor IDE performance: reduce memory usage, speed up indexing, tune AI features, and manage extensions for large codebases. Triggers on "cursor performance", "cursor slow", "cursor optimization", "cursor memory", "speed up cursor", "cursor lag".