本项目配合NICAR 2026会议，展示如何使用AI“技能”——这些模块化、自包含的指令包，扩展了Claude/Codex在特定领域任务中的能力。

该仓库包含六项技能：fecfile用于分析FEC竞选财务申报（来自Matt Hodges）、weather-forecast用于获取7天预测、image-rotator用于轮换图像、skill-creator用于构建新技能、state-county-rankings用于从地方CSV排名县人口统计，以及majority-minority-change用于人口普查快照之间的阈值交叉分析。它们共同展示了技能的广泛功能，从领域特定数据分析到简单的工具，再到技能系统本身的元工具。

与会者快速入门

关于带有复制/粘贴命令的逐步教程，请参见：

• 01_quickstart_underlying_code.md
• 02_skills_learning_notes.md（会话笔记、练习和故障排除）
• 03_fecfile_examples.md（使用FEC申报示例）
• 04_build_a_skill_from_your_code.md（从现有的 R/Python 脚本构建技能）
• '05_skill_build_example_state_county_rankings。

NICAR 2026 Session on using “Skills”

This project accompanies the NICAR 2026 session showing how to use AI “skills”, the modular, self-contained instruction packages that extend Claude/Codex’s capabilities for domain-specific tasks.

This repo includes six skills: fecfile for analyzing FEC campaign finance filings (from Matt Hodges), weather-forecast for fetching 7-day forecasts, image-rotator for rotating images, skill-creator for building new skills, state-county-rankings for ranked county demographics from local CSVs, and majority-minority-change for threshold-crossing analysis between Census snapshots. Together they demonstrate the range of what skills can do, from domain-specific data analysis to simple utilities to meta-tooling for the skill system itself.

Attendee Quickstart

For a step-by-step tutorial with copy/paste commands, see:

• 01_quickstart_underlying_code.md
• 02_skills_learning_notes.md (session notes, exercises, and troubleshooting)
• 03_fecfile_examples.md (worked FEC filing examples)
• 04_build_a_skill_from_your_code.md (build a skill from existing R/Python scripts)
• 05_skill_build_example_state_county_rankings.md (rank counties within a state from local demographics CSVs)
• 06_skill_build_example_majority_minority_change.md (analyze threshold crossings between two Census snapshots)

What is a skill?

When you ask an AI assistant to help with data analysis, it has to figure out on the fly how to approach the problem: which fields exist, how to handle edge cases, what the right tool or library is. Without guidance, it may guess at field names, rewrite the same boilerplate code every session, or make different choices each time. A colleague asking the same question might get a different approach, different column names, different aggregation logic.

A skill solves this by giving Claude or Codex a standing set of instructions before the conversation begins, a documented, version-controlled playbook that the AI follows every time. A skill is just a folder containing a SKILL.md file (plain Markdown with a short YAML header) plus any helper scripts or reference documents the AI should use. You write it once, commit it to your repo, and every session (and every collaborator) gets the same behavior.

How skills get invoked

Skills can trigger in two ways: