Cornell AI Hub: Claude Code Skills

A small collection of Claude Code skills published by the Cornell AI Hub. Skills here are designed to work with the Cornell AI Gateway (LiteLLM-fronted) and follow Anthropic's Agent Skills format.

What is a skill?

A skill is a folder containing a SKILL.md file with YAML frontmatter:

---
name: my-skill
description: One-line description used by Claude to decide when to invoke this skill.
---

# Skill body

Step-by-step instructions Claude follows when the skill is invoked...

When the skill folder is placed in a skills directory Claude Code knows about, the user can invoke it with /my-skill … or Claude will route to it automatically when the request matches the description.

Available skills

Install a skill (Claude Code)

Pick the install scope:

• Personal (user-level): ~/.claude/skills/
• Project-level: .claude/skills/ in your repo

Full clone (works for every skill, including multi-file ones)

SKILL=gateway-usage   # swap in any skill folder name from the table above
git clone https://github.com/cu-aaii/claude-skills.git /tmp/cu-claude-skills
cp -r "/tmp/cu-claude-skills/$SKILL" ~/.claude/skills/

Sparse grab without cloning the whole repo

This only works for single-file skills that ship nothing but SKILL.md (e.g. gemini-search):

mkdir -p ~/.claude/skills/gemini-search
curl -sSL https://raw.githubusercontent.com/cu-aaii/claude-skills/main/gemini-search/SKILL.md \
  -o ~/.claude/skills/gemini-search/SKILL.md

Skills with extra files under scripts/ or reference/ (e.g. gateway-usage) need every file grabbed explicitly, or use the full-clone method above instead:

mkdir -p ~/.claude/skills/gateway-usage/scripts
curl -sSL https://raw.githubusercontent.com/cu-aaii/claude-skills/main/gateway-usage/SKILL.md \
  -o ~/.claude/skills/gateway-usage/SKILL.md
curl -sSL https://raw.githubusercontent.com/cu-aaii/claude-skills/main/gateway-usage/scripts/usage.sh \
  -o ~/.claude/skills/gateway-usage/scripts/usage.sh
chmod +x ~/.claude/skills/gateway-usage/scripts/usage.sh

Restart Claude Code (or run /skills to confirm it loaded).

Installing in the Claude Code Desktop app

The Desktop app's in-session agent runs shell commands inside its own sandbox, which on some setups blocks mkdir/cp/chmod writes directly into .claude/skills/ even though the directory exists. Two ways around it:

1. Preferred: run the install commands above in a regular system terminal (macOS Terminal, iTerm, PowerShell, or Git Bash on Windows), not inside the Desktop app's own agent session. Then restart the Desktop app, or run /skills, to pick up the new skill.
2. If you only have the in-app agent available: ask it to fetch the files (via curl or git clone into a normal, non-.claude working directory, which is not sandboxed) and then write them into .claude/skills/ using its file-write tool rather than shell cp/mkdir. If a shell command into .claude/skills/ fails with "Operation not permitted," that is this sandbox, not a broken install; switching to a direct file write resolves it.

Windows notes

Skills with a bundled bash script (like gateway-usage) need a bash-compatible shell. Claude Code's Bash tool already provides one on Windows (Git Bash or WSL under the hood), so no extra setup is required to run a skill's script from within Claude Code. If you want to run scripts/usage.sh directly from a native terminal outside Claude Code, use Git Bash or WSL, not cmd.exe or plain PowerShell.

Credential resolution differs slightly on Windows: macOS Keychain lookup is skipped automatically (there is no Windows equivalent wired up yet), so Windows users should set ANTHROPIC_AUTH_TOKEN in the environment, or rely on the cornell-ai-helper config.toml fallback, to supply the gateway key.

Cornell prerequisites

Most skills in this repo call the Cornell AI Gateway (https://api.ai.it.cornell.edu) and require:

• A Cornell AI Gateway key. Set as ANTHROPIC_AUTH_TOKEN in the environment Claude Code runs in.
• CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 (required for the LiteLLM-fronted gateway).
• Cornell network or VPN access if any skill ever calls campus-only endpoints (none currently).

The gateway uses Authorization: Bearer <key>, not x-api-key. Model names accept both Cornell-prefixed forms (anthropic.claude-4.6-sonnet) and native names (claude-opus-4-7). Some skills use Cornell-specific model aliases (e.g., google-enterprise-web-search); see each skill's SKILL.md.

Cornell gateway docs: see your local Cornell/Documentation/AI-API-GATEWAY.md if you have access, or the AI Hub team for onboarding.

Contributing

See CONTRIBUTING.md for skill format expectations, naming conventions, and the proposal flow.

If you have an idea for a new skill, open an issue using the New skill proposal template before opening a PR.

License

MIT. Skills here are intended to be useful to other Claude Code users; the Cornell-gateway specifics are configurable.

Maintainers

• Cornell AI Hub
• Skill authors are credited in each skill's SKILL.md body where relevant.