Evals
Skills for building LLM evals. Starts with interactive error discovery: build a review app, sample diverse traces, and organize human annotations into failur...
Evals is an open-source Claude Code plugin in Productivity & Automation. It bundles 9 skills, 0 slash commands, and 0 subagents to extend Claude Code with verified install commands and live GitHub tracking.
How to Install & Test
Official Claude Code installation commands generated from verified marketplace manifests.
Standard Verification
Plugins can run code on your computer through hooks, commands, and MCP servers. Read what a plugin contains and only install ones you trust.
What this plugin can run:
- 9 contextual agent skill(s)
Installation Time: Running Claude Code marketplace commands resolves dependencies, AST verification, and hooks. This usually takes 10–30 seconds. Wait for the CLI confirmation before proceeding.
Run these slash commands inside your Claude Code interactive session:
Registers the author's plugin catalog with your Claude Code client:
/plugin marketplace add ai-evals-course/evals-skills
Installs the package into your chosen scope (user, project, or local):
/plugin install evals@ai-evals-course
Reloads active plugins in your session without exiting:
/reload-plugins
Test or inspect the plugin locally using official Claude Code CLI commands:
git clone https://github.com/ai-evals-course/evals-skills.git
Loads the local plugin into this terminal session only:
claude --plugin-dir ./
claude plugin validate ./
claude plugin install ./
Copy this audit-and-install prompt and paste it into Claude Code or your AI coding assistant:
Please inspect the Claude Code plugin at https://github.com/ai-evals-course/evals-skills. 1. Read the manifest (.claude-plugin/plugin.json or marketplace.json) and scan components (commands, skills, agents, hooks, and .mcp.json). 2. Summarize every hook lifecycle event, script, command, and external MCP server it defines. 3. Verify whether it requests API keys or accesses sensitive local paths. 4. Ask for my confirmation. Once approved, run: /plugin marketplace add ai-evals-course/evals-skills /plugin install evals@ai-evals-course /reload-plugins
What's Inside This Package
Components declared in the plugin tree: skills, slash commands, dedicated subagents, lifecycle hooks, and MCP servers.
Overview & Documentation
Evals — Skills for building LLM evals. Starts with interactive error discovery: build a review app, sample diverse traces, and organize human annotations into failure modes.
Includes 9 skills.
Install in Claude Code: /plugin marketplace add ai-evals-course/evals-skills, then /plugin install evals@ai-evals-course.
Source: ai-evals-course/evals-skills (1808 GitHub stars as of 2026-11-18).
Frequently Asked Questions
Answers to common questions regarding Evals compatibility, safety, and configuration.
Frequently Asked Questions
Direct answers to common questions about using these skills and tools.
How do I install Evals in Claude Code?
Add the marketplace via "/plugin marketplace add ai-evals-course/evals-skills", then run "/plugin install evals@ai-evals-course". Finally, run "/reload-plugins" to activate the package in your current session.
What capabilities and components are included inside Evals?
Evals bundles 9 contextual skills, 0 custom commands, 0 subagents, no event hooks, and 0 external MCP server connections.
Is Evals safe to run on my machine?
This package is classified as "low" risk. It runs locally without unauthorized elevation. Always inspect the code before installing.
How do I test Evals locally without permanent installation?
Clone the repository and run "claude --plugin-dir ./". This loads the plugin strictly for the active terminal session without changing your global settings.