數據與 AI

Guide for authoring, designing, building, writing, and developing specialized AI agents. Use when creating, updating, reviewing, or improving agents that handle specific tasks with focused expertise. Helps design AGENT.md files, choose models (Sonnet/Haiku/Opus), define focus areas, configure tool restrictions, and decide between agents, skills, and commands. Expert in agent validation, best practices, and troubleshooting.

Production-ready Celery task templates with error handling, retries, rate limiting, time limits, and custom task classes. Use when creating Celery tasks, implementing retry logic, adding rate limiting, setting time limits, building custom task classes, validating task inputs with Pydantic, handling database operations, making API calls, or when user mentions task patterns, retry mechanisms, task templates, error handling, task best practices.

Analyzes HTML/JSX/TSX files for SEO and accessibility issues including WCAG 2.1 AA compliance, color contrast (4.5:1), heading hierarchy, meta tags, image alt text, and ARIA attributes. Use when checking web pages for SEO, accessibility, WCAG compliance, or when user mentions "a11y", "contrast", "alt text", "meta tags", "heading structure", or "accessibility audit".

Farm out AI tasks via Named Pipe to get isolated context. Use when processing large files, batch summarization, or to avoid context pollution. Each call runs in fresh context.

Extracts, transforms, and analyzes NBA statistics using the nba_api Python library. Use when working with NBA player stats, team data, game logs, shot charts, league statistics, or any NBA-related data engineering tasks. Supports both stats.nba.com endpoints and static player/team lookups.

Use this skill when troubleshooting OpenStudio or EnergyPlus energy models that fail to simulate, have geometry errors (intersecting surfaces, non-planar surfaces, story organization issues), need HVAC validation against Engineer of Record specifications, require LEED Appendix G baseline generation, or need systematic diagnostics for complex commercial building models. Handles model triage, geometry rebuild decisions, EOR specification mapping, and quality assurance for LEED compliance.

Analyze and optimize database schemas and relationships

Decomposes a decision-complete PRD into a clear, minimal set of user-meaningful features with explicit dependencies, producing a roadmap-ready structure without prioritization, scheduling, or implementation detail.

Find Mozilla telemetry probes and Glean metrics. Use when user asks about: Firefox metrics, Glean probes, telemetry data, accessibility probes, search metrics, or any Mozilla product instrumentation.

Deep dive into memory layout, including strides, C vs Fortran order, and zero-copy view generation via stride tricks. Triggers: strides, C-order, Fortran-order, memory locality, stride_tricks.

Step-by-step procedural guidance for executing the /plan command workflow. Covers Phase 0 (Research), Phase 1 (Design), and Phase 2 (Validation) for implementation planning with constitutional compliance.This skill provides step-by-step procedural guidance for tasks handled bythe planning-agent agent. Use when you need to understand the workflow orexecute tasks that planning-agent would typically handle.Triggered by: [TODO: Add trigger phrases]

GPU monitoring and performance metrics for Ollama inference. Check GPUstatus, VRAM usage, loaded models, and inference performance metricslike tokens per second.

Use when starting a new feature that needs planning - runs brainstorming then creates numbered task folder in _tasks/

Synchronizes Cursor Rules (.mdc files in .cursor/rules/) to CLAUDE.md by generating a Rules section with context-efficient descriptions and usage instructions. Use when setting up Cursor Rules for the first time, after adding or modifying rules, or when the Rules section in CLAUDE.md is missing or outdated.

Instructions for downloading the ai4space repository, useful when the user wants to start working with the lunar base model.

Use when Codex needs semantic web search, news reconnaissance, or citation gathering through the Exa MCP server.

无头模式 AI CLI 调用技能：支持 Gemini/Claude/Codex CLI 的无交互批量调用，包含 YOLO 模式和安全模式。用于批量翻译、代码审查、多模型编排等场景。

Comprehensive guide to phylogenetic tree visualization including matplotlib (Python), ggtree (R), and publication-quality figure generation. Covers rectangular, circular, radial layouts, bootstrap annotations, branch coloring, clade highlighting, and interactive visualizations. Based on best practices from Yulab's Tree Data Book.

Work with markdown-based task lists in .llm/todo.md files. Use when managing tasks, working with todo lists, extracting incomplete tasks, marking tasks complete, or implementing tasks from a task list.

Extract lessons from task failures using Reflector agent, merge insights via Curator pattern, achieving 10% accuracy improvement without context collapse. Use for post-failure analysis, lesson extraction, self-improvement, and building resilience. Analyzes what went wrong and formulates actionable lessons. Triggers on "learn from failure", "analyze mistake", "reflect on error", "post-mortem", "failure analysis", "what went wrong".