Share AI agent context across tools and teams.
dotai takes canonical context files — skills, prompts, agent definitions, and instructions — and installs them into the config directories of supported AI coding agents. Write once, distribute everywhere. Your team gets consistent AI behavior across Copilot, Claude Code, Cursor, and more.
Requires Node.js 18+ (or Bun/Deno).
Supports OpenCode, Claude Code, Codex, Cursor, and GitHub Copilot.
npx dotai add owner/repo # install context from any GitHub repo
npx dotai add owner/repo --dry-run # preview without writing filesAI coding agents each store context in different places with different formats. Keeping instructions, prompts, and skills in sync across agents is manual and error-prone.
dotai solves this with canonical authoring: write a single
PROMPT.md, AGENT.md, or INSTRUCTIONS.md and dotai transpiles it into every
target agent's native format automatically.
- Write once — one canonical file fans out to all targets
- 5 targets — Copilot, Claude Code, Cursor, Codex, OpenCode
- Team sharing —
npx dotai add owner/repogives every teammate the same context - No lock-in — canonical files are plain markdown with YAML frontmatter
npm install -g dotai # install globally for regular use
npx dotai # run without installing# Add context from a GitHub repo
npx dotai add owner/repo
# Limit to specific targets
npx dotai add owner/repo --targets copilot,claude,cursor
# Install specific instructions or skills
npx dotai add owner/repo --instruction my-instructions --skill db-migratedotai discovers skills, prompts, agent definitions, and instructions in the source repo and transpiles them for your selected targets.
| Layer | Canonical file | Install behavior |
|---|---|---|
| Skills | SKILL.md |
Passthrough (symlink or copy) |
| Prompts | PROMPT.md |
Transpile per supported target |
| Agents | AGENT.md |
Transpile per supported target |
| Instructions | INSTRUCTIONS.md |
Append per target |
See Source repo layout for where to place these files in your repo so dotai discovers them.
dotai accepts sources in many formats:
npx dotai add vercel-labs/agent-skills # GitHub shorthand
npx dotai add https://github.com/vercel-labs/agent-skills # full URL
npx dotai add https://gitlab.com/org/repo # GitLab
npx dotai add git@github.com:org/repo.git # any git URL
npx dotai add ./my-local-context # local path| Command | Description |
|---|---|
add <package> |
Discover, select, transpile, and install context |
remove [name] |
Remove installed context |
list |
List installed items |
find [query] |
Search for skills & preview all context in a repo |
check |
Check for available updates |
update |
Update installed items to latest versions |
init [name] |
Create a context template (skill, prompt, agent, instruction) |
restore |
Restore from lock files |
Transpilation support by agent
| Agent | Skills | Prompts | Agents | Instructions |
|---|---|---|---|---|
| GitHub Copilot | ✅ | ✅ | ✅ | ✅ |
| Claude Code | ✅ | ✅ | ✅ | ✅ |
| OpenCode | ✅ | ✅ | ✅ | ✅ |
| Cursor | ✅ | .github/agents) |
✅ | |
| Codex | ✅ | — | — | — |
- Cursor prompts: Cursor reads Copilot's
.github/prompts/path. CanonicalPROMPT.mdis not transpiled to a Cursor-specific format. - Cursor agents: Cursor reads
.github/agents/from the Copilot path. CanonicalAGENT.mdtranspiles to Copilot format, which Cursor picks up. - Instructions:
INSTRUCTIONS.mdcontent is appended as marker-delimited sections to project-wide files (CLAUDE.md,AGENTS.md,.github/copilot-instructions.md). Cursor and OpenCode shareAGENTS.md.
Skill installs target 5 targets.
- CLI Reference — all commands, flags, options, and authoring format
- Supported Targets — full list of skill install targets
dotai started as a fork of vercel-labs/skills / skills.sh. The inherited skills install pipeline remains first-class. dotai extends it with transpilation of prompts, agent definitions, and instructions to multiple targets.
- vercel-labs/skills — upstream fork source
- skills.sh — skill discovery and search
- Agent Skills specification
MIT