Getting started · RC4.1

Your first collection.

Kitrove collects skills from your agent tools and helps you apply compatible capabilities elsewhere. You choose what to adopt, where to put it, and which changes to approve.

Install the preview

On an Apple Silicon Mac with Homebrew:

brew install --cask kapital-labs/tap/kitrove-rc

For an existing preview installation:

brew update
brew upgrade --cask kapital-labs/tap/kitrove-rc

Homebrew owns the binary. Uninstalling the cask preserves your collection and local state. Do not use the standalone installer to replace Homebrew-managed files.

Windows and Linux package-manager installs are not published. Intel Mac, Windows x86_64, and Linux x86_64 users can consult the manual release and verification guide. Authenticate a download before its first execution.

Set up your first computer

kitrove init

1. Choose Set up my first computer.
2. Choose a collection folder. Kitrove suggests ~/Kitrove and asks for a name for this computer. Review the collection and machine-local state locations before confirming creation.
3. Pick skills by the numbers in the scan results, or choose all. Review the source tool as well as the name. The same directory can appear through more than one tool.
4. Review and confirm each adoption. Kitrove leaves the source files in place.
5. Choose Apply selected items, pick destinations, then review and confirm the separate application plan.

Enter skips a confirmation. Type q to stop. Steps you already confirmed stay completed; the walkthrough is not one all-or-nothing transaction.

Your collection and local state are separate. On macOS, the default state folder is ~/Library/Application Support/kitrove. Setup shows both paths and does not edit your shell profile.

Return when you want to change something

kitrove configure

Select your existing collection to collect more skills, apply another selection, sync, or import selected items. To choose a non-default collection:

kitrove configure --environment /absolute/path/to/collection

Nothing starts syncing in the background. You choose a backend for each operation, review the plan, and confirm it.

Bring your collection to a second computer

On the first computer, open Configure and choose Connect / sync the whole collection. Use an existing filesystem/shared directory, HTTPS Git remote, or SSH Git remote with an explicit known-hosts file. Create and authorize a remote repository separately.

This must be a Kitrove snapshot backend, not a regular checkout of skill files. An empty existing backend can receive the first approved publication. Review changes on both sides: full sync is bidirectional.

Install Kitrove on the second computer and run kitrove init. Choose the route that fits:

• Connect a second computer to the whole collection. Create a new local identity and review full sync. Your agent files stay untouched until you choose Apply.
• Combine this computer's skills with an existing collection. Connect an empty collection first to establish shared history. Then collect this computer's skills and review publishing the additions.
• Import specific items. Choose assets or packs for a one-time import. Packs include their required members. This preserves unrelated local items and does not change the remote.

Two independently populated collections without shared history cannot full-sync automatically. Use selective import for chosen items. Do not delete local state to force a connection.

Conflicting content under the same identity stops the operation for review. RC4.1 does not choose an automatic winner or create a keep-both rewrite.

Inspect skills and choose by name

kitrove scan
kitrove scan --harness claude --scope user
kitrove adopt humanizer --harness claude --scope user

The last command is an example for a uniquely named skill called humanizer. Replace it with a name from your scan. Ambiguous names need a narrower source or scope, or the exact observation ID from kitrove scan --verbose.

Scanning does not change files. Adoption still captures fresh content and requires a confirmed plan. For automation, use kitrove scan --json; bare Init and Configure require a terminal.

Know the boundaries

• Collection sync excludes credentials, trust decisions, machine paths, local bindings, and deployment receipts. Install and authorize each agent tool on each computer.
• Kitrove reports transfer fidelity. A successful import does not mean identical behavior in another tool.
• Executable capabilities need separate local trust. Setup does not grant it or install prerequisites. Pi and OpenCode may ask for an independently trusted local binary for a bounded version check.
• Kitrove refuses unmanaged destination conflicts instead of overwriting existing files.
• Complete pending RC4.1 application transactions before downgrading to RC3, which does not understand the new nested-home transaction layout.

RC4.1 remains a prerelease. Tests on the existing Apple Silicon signing host do not establish clean-Mac quarantined online/offline first launch, other native target acceptance, or real two-machine product acceptance.

Go deeper or get help

The repository contains the full setup guide, release verification instructions, and acceptance checklist.

Report problems in GitHub issues. Include the version and command, but remove credentials, private paths, and sensitive skill content from logs before sharing.