數據與 AI

CRITICAL - Guide for using Claudish CLI ONLY through sub-agents to run Claude Code with OpenRouter models (Grok, GPT-5, Gemini, MiniMax). NEVER run Claudish directly in main context unless user explicitly requests it. Use when user mentions external AI models, Claudish, OpenRouter, or alternative models. Includes mandatory sub-agent delegation patterns, agent selection guide, file-based instructions, and strict rules to prevent context window pollution.

Implements UI components from scratch based on design references (Figma, screenshots, mockups) with intelligent validation and adaptive agent switching. Use when user provides a design and wants pixel-perfect UI implementation with design fidelity validation. Triggers automatically when user mentions Figma links, design screenshots, or wants to implement UI from designs.

Comprehensive PDF manipulation toolkit for extracting text and tables, creating new PDFs, merging/splitting documents, and handling forms. When Claude needs to fill in a PDF form or programmatically process, generate, or analyze PDF documents at scale.

⚡ PRIMARY TOOL for: 'what's tested', 'find test coverage', 'audit test quality', 'missing tests', 'edge cases', 'test patterns'. Uses claudemem v0.3.0 AST with callers analysis for test discovery. GREP/FIND/GLOB ARE FORBIDDEN.

Run multiple AI models in parallel for 3-5x speedup with ENFORCED performance statistics tracking. Use when validating with Grok, Gemini, GPT-5, DeepSeek, or Claudish proxy for code review, consensus analysis, or multi-expert validation. NEW in v3.1.0 - SubagentStop hook enforces statistics collection, MANDATORY checklist prevents incomplete reviews, timing instrumentation examples. 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", "claudish", "multiple models", "parallel review", "external AI", "consensus", "multi-model", "model performance", "statistics", "free models".

Common agent patterns and templates for Claude Code. Use when implementing agents to follow proven patterns for proxy mode, TodoWrite integration, and quality checks.

Memory cleanup and organization skill for AI employees. Provides guidelines for detecting duplicates, fixing misclassified files, and removing stale information from memory storage.

Coordinate multiple agents in parallel or sequential workflows. Use when running agents simultaneously, delegating to sub-agents, switching between specialized agents, or managing agent selection. Trigger keywords - "parallel agents", "sequential workflow", "delegate", "multi-agent", "sub-agent", "agent switching", "task decomposition".

⚡ PRIMARY TOOL for: 'comprehensive audit', 'deep analysis', 'full codebase review', 'multi-perspective investigation', 'complex questions'. Combines ALL detective perspectives (architect+developer+tester+debugger). Uses Opus model. REPLACES grep/glob entirely. Uses claudemem v0.3.0 AST with ALL commands (map, symbol, callers, callees, context). GREP/FIND/GLOB ARE FORBIDDEN.

Extract actionable tasks assigned to the user from email text. Filters out informational emails (announcements, newsletters, ads, automated reports) and only processes emails with clear action requests. Handles group emails by identifying user-specific assignments.

Common patterns for extracting analytics data from GA4 and GSC with API handling

Core principles and project structure for React 19 SPA development. Covers stack overview, project organization, agent execution rules, and authoritative sources. Use when planning new projects, onboarding, or reviewing architectural decisions.

Analyze bank transaction CSVs to find recurring charges, categorize subscriptions, and cancel what you don't need. Use when user says "cancel subscriptions", "audit subscriptions", "find recurring charges", or "what am I paying for". Supports Apple Card, Chase, Mint, and generic CSV formats. Outputs interactive HTML audit with copy-to-cancel workflow.

Use when user has complex multi-agent workflows, needs to coordinate sequential or parallel agent execution, wants workflow visualization and control, or mentions automating repetitive multi-agent processes - guides discovery and usage of the orchestration system

Create and execute temporary scripts (Python, Node.js, shell) during workflow execution for API integrations, data processing, and custom tools. Use when user needs to interact with external APIs, process data with specific libraries, or create temporary executable code.

Use when user says "create workflow", "create a workflow", "design workflow", "orchestrate", "automate multiple steps", "coordinate agents", "multi-agent workflow". Creates orchestration workflows from natural language using Socratic questioning to plan multi-agent workflows with visualization.

Debug workflow execution issues including syntax errors, agent failures, variable problems, and execution errors. Use when workflows fail, produce unexpected results, or user asks for debugging help.

Manages temporary and defined agents including creation, promotion, cleanup, and namespacing. Use when user creates custom agents, asks about agent lifecycle, temp agents, or agent management.

Use and customize workflow templates for common scenarios. Use when user wants to use a template, asks about available templates, or wants to customize existing workflows.

Design custom syntax elements with reuse-first approach for workflow orchestration. Use when user needs custom operators, checkpoints, or syntax patterns not available in core syntax.