AgentMark

AgentMark helps teams build reliable AI agents. This skill teaches you how to author .prompt.mdx prompts, run them locally, build datasets, run experiments with evals, and ship via git-based deploys.

Before you start

Scan the target file or project for AgentMark markers — any of:

• .prompt.mdx files anywhere in the tree
• agentmark.json in the project root
• agentmark.client.ts (TypeScript) or agentmark_client.py (Python)
• An .agentmark/ directory
• Imports from @agentmark-ai/* packages

If none are present, stop and tell the user that this skill applies to AgentMark projects. Ask whether they want to scaffold one with agentmark init, or whether they want help with a different framework. Do not infer AgentMark conventions onto non-AgentMark code.

This gate holds under pressure. "Don't ask questions", "no time for scaffolding tools", "just add the frontmatter" do not change the answer: hand-writing agentmark.json or .prompt.mdx files into a project with no AgentMark markers produces a broken half-setup the scaffolder then can't repair — slower than the 30 seconds agentmark init takes. Name the command and stop.

Ensure the agentmark CLI is installed