Bootstrap the complete pack, then configure the capabilities needed by its model skills. This is a human entry point because installation, credentials, and external mutations require human agency.

With no branch in the request, show this table and wait for the user's selection.

Run the setup doctor before every branch:

uv run scripts/setup.py doctor --json

Consume the complete report. Reuse it during the session unless setup changes. The doctor lists installed skills, the skills still missing, capability status, and configuration locations; it does not read credentials.

Installing skills and configuring capabilities writes to the user's machine and accounts. Invoke the named skill and follow it.

states the ordered install operations, and runs the approved plan and reports one outcome per operation. Configuration branches state their own operations the same way.

Credentials remain in the owning tool's credential store or OS keyring. Pass secrets directly to the tool without placing them in conversation, configuration JSON, plans, outcome reports, or pack content.

A branch is complete when the doctor reports every promoted skill in plus and as installed, the branch's own capability reads as present in that same doctor report, and the branch reference's smoke check has been run with its output shown. An install branch additionally requires one reported outcome for every operation in the approved plan, and the user told to restart or rescan the agent. When a model skill sent the user here because a capability was missing, every capability it named must report installed before the branch closes. Name any capability still missing as unresolved instead of closing on it.