Skip to main content
The Galtea Agent Skill is an open-source Agent Skill that teaches AI coding assistants — Claude Code, Cursor, Windsurf, and other Agent-Skills-compatible tools — how to work with Galtea. With the skill installed, your agent authenticates, discovers the right endpoints, runs evaluations, and inspects sessions and traces without you having to explain Galtea first. Source repository: Galtea-AI/skills

​
Why use it

Without the skill, the agent improvises: it guesses endpoints, re-reads the docs from scratch, and often picks a slower path than the one Galtea recommends. With the skill loaded, the agent follows the workflows our team has already validated and sticks to the conventions encoded in the skill itself.

​
Install

Install the Galtea Agent Skill to let your coding agent access all Galtea features.
Ask your coding agent to install the skill by pointing to the GitHub repository:
Install the Galtea Agent Skill from github.com/Galtea-AI/skills.
The agent reads the repo, picks the installation path for its own runtime (Claude Code, Cursor, Windsurf, etc.), and wires the skill up.

​
Prerequisites

You need a Galtea account and an API key: 
export GALTEA_API_KEY=gsk_...
API keys live in the Galtea dashboard under Settings → API Keys. Each account has a single key; regenerating permanently replaces it.
If GALTEA_API_KEY is unset, the skill prompts the agent to walk you through a one-time paste flow and caches the key at ~/.galtea/api-key with mode 600.

​
Usage examples

Once installed, the agent automatically picks up the skill when the task involves Galtea. Try prompts like:
  • “Set up a new product version and connect my endpoint.”
  • “Run an evaluation for version <id> and show me the failures.”
  • “List the metrics linked to this product’s specifications.”
  • “Why did session <id> end up in PENDING_HUMAN?”
  • “Create a test for the spec I just drafted and kick off evaluations.”
The skill handles authentication, calls the right endpoints, polls async evaluations until they settle, and surfaces the results.

​
Feedback

Something not working as expected, or want the skill to cover a new workflow? Open an issue on the Galtea-AI/skills repository — the agent can also do this for you via the built-in skill-feedback flow.

Quickstart

Zero-to-first-evaluation walkthrough.

Python SDK

Prefer Python? Install the official SDK.