The aidev CLI

A package manager for AI development building blocks. Humans and coding agents use the same commands.

Install

npm install -g aidev
aidev login
aidev search "react migration"
aidev install react-migration-agent

Where files go

Each provider adapter translates one neutral artifact into that tool’s current format. Nothing is overwritten silently: conflicts abort, --force backs up first, and every install is a rollback point.

Artifact	GitHub Copilot	Claude Code	Codex
skill	`.github/skills/<n>/SKILL.md`	`.claude/skills/<n>/SKILL.md`	`.agents/skills/<n>/SKILL.md`
agent	`.github/agents/<n>.agent.md`	`.claude/agents/<n>.md`	installed as a skill (no agent file format)
prompt	`.github/prompts/<n>.prompt.md`	user-invoked skill	skill
instruction	`.github/instructions/<n>.instructions.md`	block in `CLAUDE.md`	block in `AGENTS.md`
mcp	`.vscode/mcp.json`	`.mcp.json`	`.codex/config.toml` or `~/.codex/config.toml`

For coding agents

Every command supports --json (a single JSON document on stdout), --yes (non-interactive consent) and --no-color. Authenticate with AIDEV_TOKEN.

aidev search "react 19 migration" --json
aidev info react-19-migration --json
aidev install react-19-migration --yes --json --provider copilot

Exit	Meaning
0	success
2	usage error / provider not determinable
3	authentication required
4	not found
5	conflict (would overwrite existing config)
6	blocked by company policy
7	consent required (re-run with `--yes`)
8	validation failed
9	network error

Safety

Artifacts are untrusted. Installing never executes anything: scripts are copied, not run. Permissions are shown before install, high-severity scan findings require --allow-risky, content hashes are verified, and company-blocked artifacts are refused.

Publish

aidev init artifact skill my-skill
cd my-skill && aidev publish --dry-run
aidev publish --org acme --visibility organization