AI Agents
Give ChatGPT, Cursor, Codex, OpenCode, Antigravity, T3 Chat, and other agents clean Expojet context.
Every documentation page has Copy Markdown and an Open menu. The menu opens supported web agents directly, while Copy Markdown is the universal path for local and desktop agents such as Codex, Cursor, OpenCode, and Antigravity.
Install stack-aware skills
After creating a project, interactive setup offers Install recommended, Customize, or Skip (the default). Customize lets you toggle each applicable Expo, authentication, backend, database/ORM, and web skill individually. Choose your coding agents and then Project (recommended) or Global scope. Review the pinned sources and destinations before confirming.
npx create-expojet@latest my-app --yes --skills --skill-agents codex --skills-scope project
npx create-expojet@latest my-app --yes --skills --skill-agents claude-code --skills-scope global
npx create-expojet@latest my-app --yes --no-skills--yes by itself does not install skills. Explicit noninteractive installation
requires agent names and confirmation through --yes. These options are separate
from the generation input and do not change presets or the app manifest.
For an existing generated project, run this from the project or a subdirectory:
expojet skills install
expojet skills install --yes --agents codex --scope global
expojet skills install --dry-run
expojet skills install --yes --agents codex --skill expo-overview expo-routerDry runs do not download sources or write files. --skill accepts applicable
catalog IDs, shown in the preview. Omit it to use recommendations or customize
interactively. Reload your agent if newly installed skills are not discovered.
Coverage and compatibility
Every stack gets selectable Expo fundamentals. Expo Router, Expo UI for Liquid Glass, and EAS skills appear only when relevant. Reviewed sources also cover Clerk, Hono, Prisma, Neon, Supabase Postgres, Next.js, and Convex. Convex guidance comes from a clearly labeled community source; the other sources are official.
Some integrations do not yet have reviewed, redistributable catalog coverage, including Better Auth, Express, NestJS, Drizzle, and styling/state/analytics integrations. Setup reports these gaps rather than substituting unrelated skills. Skill guidance is not an SDK compatibility certificate: upstream instructions may target different dependency versions, including Clerk guidance. Check your installed SDK and package versions before applying an agent's suggestions.
Supported targets are Codex, Claude Code, Cursor, GitHub Copilot, Gemini CLI,
OpenCode, Antigravity, Windsurf, and a shared directory. Project installs use
.agents/skills, except Claude Code (.claude/skills) and Windsurf
(.windsurf/skills). Global destinations follow each target's configuration and
are printed before installation. Agent versions must support skill discovery at
those locations; installation does not configure an agent or MCP server.
Safe retries
Sources and the installer are pinned. Skills are copied, not symlinked; included scripts are never run. Identical skills are skipped. Different existing content is preserved and reported as a conflict—there is no automatic overwrite or update.
Project installs record successful destinations in expojet-skills.lock.json.
Global installs keep that record in Expojet's user-data directory (normally
~/.expojet, or EXPOJET_CONFIG_DIR) without changing project files.
Failures are reported per source and independent sources continue. Cancellation preserves successful installations. Post-create skills failure leaves app creation successful; incomplete standalone installation returns a nonzero exit code. Use the scoped retry command printed by setup after fixing connectivity or reviewing existing content. Each external command has a two-minute timeout.
Machine-readable endpoints
| Endpoint | Best for |
|---|---|
/llms.txt | A compact index that helps an agent find the right page |
/llms-full.txt | The complete documentation in one request |
/llms.mdx/docs/<path>/content.md | Clean Markdown for one documentation page |
Use the smallest source that contains the needed context. A focused page is faster and less likely to distract an agent than the full documentation dump.
Example prompt
Read https://www.expojet.dev/llms.txt, then use only the relevant linked Expojet
documentation to generate a project. Do not expose server secrets
to the mobile bundle, and verify the result with expojet doctor.Replace the production host with http://localhost:3000 while developing this
site locally.
Agent compatibility
- ChatGPT and other supported web agents: choose the agent from Open.
- Cursor, Codex, OpenCode, and Antigravity: copy the page Markdown or point
the agent at its
content.mdURL. - T3 Chat and other chat clients: paste the copied Markdown with your
request, or attach
llms.txtwhen URL access is available.
No proprietary integration is required. These endpoints are plain text and Markdown, so they remain portable across current and future AI tools.