oo-init

name: oo-init description: Use when the user wants to start, scaffold, or create their oo publishable identity. Triggers on phrases like "init my agent", "set up oo", "make my Claude/Codex setup publishable", "create agent.json", "start a new oo bundle". allowed-tools: Bash, Read

Initialize the user's publishable identity

There is no bundle directory. The user's ~/.co/ IS the bundle:

~/.co/keys/agent.key      identity
~/.co/keys.env            managed-key auth
~/.co/agent.json          publishable profile (alias, bio, version)
~/.co/skills/             skill library (sources from claude/codex/cursor/kiro)
~/.co/skills/index.json   discovery cache

All four are created and refreshed by one command: co setup. This skill just gathers the right inputs and shells out.

Prerequisites

co setup --help >/dev/null && co skills --help >/dev/null \
  || { echo "MISSING: pip install -U connectonion"; exit 1; }

Phase A — Gather inputs

Ask the user for two things (or infer + confirm):

1. Alias / name — default to git config user.name | tr '[:upper:]' '[:lower:]' | tr -d ' ' or $USER. Must be lowercase, alphanumeric + hyphens.
2. One-line bio — push back on placeholders ("my agent"). A vague bio is the most common reason a publication gets ignored.

Mention but don't block on co auth — publishing works without managed-key auth; only the co/* models require it.

Phase B — Run co setup

co setup --name <alias> --bio "<one-line bio>"

That single command:

• Bootstraps ~/.co/keys/agent.key (via co init in a tmpdir) if missing.
• Writes ~/.co/agent.json with the signing address, alias, bio, and skill metadata. Skips profile creation if a profile already exists unless you pass --force (which backs up to agent.json.bak).
• Runs co skills discover && co skills copy --all && co skills manifest to populate the library and merge skill metadata into agent.json. Idempotent — existing skills are not overwritten without --force. New skill entries default to publish: false.
• Reports identity, auth status, and library counts.

If ~/.co/agent.json already exists with a different alias/bio, surface the conflict to the user before passing --force — don't clobber silently.

If they're scripting (e.g. a clean test setup), --no-skills skips the library refresh.

Phase C — Point at next steps

The CLI prints a summary. Add the publishing hint if not obvious:

✓ Setup complete. Your identity is at ~/.co/agent.json.

Next:
  • Browse your library:       co skills list
  • Edit bio if needed:        $EDITOR ~/.co/agent.json
  • Publish:                   run the oo-publish skill

Anti-patterns

• Don't mkdir a separate bundle directory. There is no bundle dir. Everything lives in ~/.co/. The "bundle" abstraction was deleted on purpose.
• Don't reimplement co setup in bash. If you find yourself writing mkdir, cat > agent.json, cp ~/.claude/skills/..., or python -c "from connectonion import address" blocks, stop. The logic lives in connectonion/cli/commands/setup_commands.py and is the single source of truth.
• Don't curate which skills get published here. co setup copies every discoverable skill into the user's library. Picking which ones go into the announced profile is oo-publish's job.
• Don't sign agent.json here. Signature is a publish-time concern.
• Don't pass --force without telling the user. A backup is made, but a noisy confirmation prevents accidental loss.