Give your agent the setup prompt
The fastest path to setting up Keystroke is giving your agent the following prompt:What your agent gets
keystroke init scaffolds a project that’s ready for a coding agent from the first commit:
| File | What it gives your agent |
|---|---|
AGENTS.md | Baseline context: the CLI, project layout, discovery, and working habits |
CLAUDE.md | Symlink to AGENTS.md so Claude Code picks it up automatically |
tsconfig.json | Committed TypeScript config (extends @keystrokehq/cli/tsconfig.json) |
package.json | Scripts for keystroke dev, build, lint, typecheck, and test |
src/ | An example agent to read and extend |
keystroke docs search and keystroke docs query — see docs for agents.
Most coding agents read AGENTS.md automatically when they open the project, so they’re oriented before you ask for anything.
Start building
With the project open in your coding agent, just describe what you want. For example:Build over MCP instead
If you’d rather build from a chat-first agent like ChatGPT or Claude — with no local install — connect the Keystroke MCP server. Your agent works in a hosted workspace with the same docs access and deploy loop, so you skipkeystroke init entirely.
Give your agent the docs
This page is often the first thing an agent reads. You can hand it Keystroke’s docs in a context-aware way:-
All the docs (llms.txt)
Point your agent at the complete, machine-readable docs index:
-
A single page as markdown
Append
.mdto any docs URL to get a clean markdown version: - In your editor Every docs page can be copied or opened directly in ChatGPT, Claude, Cursor, or an MCP client from the menu at the top of the page.