Import

Switching from another coding agent? One command brings your setup over.

/import

Run it bare and Command Code scans every supported source on your machine โ€” your home directory and the current project โ€” and imports from all of them. Or name one:

/import claude # Claude Code (also: claude-code, claudecode) /import codex # Codex /import cursor # Cursor /import pi # Pi /import opencode # OpenCode /import gemini # Gemini CLI (also: gemini-cli)

An unrecognized name is rejected with the list of valid ones โ€” nothing is written.

~/.claude/ โ” ~/.codex/ โ”‚ โ”Œ ~/.commandcode/ ~/.cursor/ โ”œโ”€โ–บ /import โ”€โ”ค skills ยท agents .../ 3 more โ”‚ โ”‚ commands ยท mcp.json <project>/ โ”˜ โ”‚ AGENTS.md โ”‚ โ”” <project>/ .commandcode/โ€ฆ AGENTS.md

Five kinds of thing, at both user scope (your home directory) and project scope (this repo), depending on where the source keeps them:

ItemLands in
Skills~/.commandcode/skills/ or <project>/.commandcode/skills/
Custom agents~/.commandcode/agents/ or <project>/.commandcode/agents/
Custom slash commands~/.commandcode/commands/ or <project>/.commandcode/commands/ (subdirectory namespacing preserved)
MCP server configs~/.commandcode/mcp.json
Memory files~/.commandcode/AGENTS.md (user) or <project>/AGENTS.md (project) โ€” see Memory

Formats that are already compatible copy over as-is: agent frontmatter (name, description, tools, model) and the custom-command markdown convention โ€” including $ARGUMENTS and positional substitution โ€” mean the same thing in Command Code. Where a source uses a different format (for example Codex's TOML agents), the item is converted on the way in.

Note

A source's memory file is renamed on import โ€” CLAUDE.md becomes AGENTS.md, Command Code's memory filename โ€” and references to the old filename inside the content are rewritten so the agent isn't pointed at a file that no longer exists.

Additive. /import never overwrites your existing Command Code configuration. Anything already present is left exactly as it is and reported as skipped.

Idempotent. Running /import again is safe โ€” nothing duplicates, nothing gets clobbered.

Fault-isolated. Each item imports independently, and each source is scanned independently. One unreadable skill, one malformed MCP entry, or one source whose config format has changed can't take the rest of the run down โ€” the failure is reported and everything else still goes through.

Every item ends in one of four states, and the summary tells you which:

StatusMeaning
importedCopied (or converted and copied) into your Command Code config
skipped-existingYou already have something at that target path
skipped-unsupportedDetected but not representable in Command Code; reported so you know it exists
failedRead or write error on that one item

/import and taste onboarding solve two different problems:

CopiesLearns
/importConfig: skills, agents, slash commands, MCP servers, memory filesโ€”
/learn-tasteโ€”Your style, from your past sessions in other agents

/import brings over the things you built. Taste onboarding learns the way you work from your history in other tools. Run both and Command Code feels like home immediately: your setup is there, and so is your style.

/import # copy the config /learn-taste # learn the style
Complete guideTaste

How taste learning works, where profiles live, and how to push, pull, and share them across projects and your team.

  • Taste โ€” learns your style from past sessions, distinct from copying config.
  • Memory โ€” where imported memory files land.
  • MCP โ€” how imported server configs are used.
  • Skills and Agents โ€” what imported skills and agents become.