Add VibeKit to an existing project
Run VibeKit from the root of the existing project:
vibekit initThe interactive setup asks which supported agent harnesses, skills, and MCP servers to configure. Read the preview before confirming changes to an existing agent-instructions file.
What it adds
Section titled “What it adds”VibeKit installs the selected skills in each harness’s discovery location, writes or updates its MCP configuration, and generates project instructions that route an agent to the right skill for the task. It preserves unrelated MCP servers in supported JSON configurations.
The VibeKit MCP entry invokes your local vibekit mcp command over stdio. The
agent receives the stock read and actions for accounts, assets, contracts,
transactions, and network data, plus the default plugins (NFD, Alpha Arcade,
Vestige, Pera, web search; Haystack swaps with a router key). It defaults to LocalNet while serving localnet, testnet, and
mainnet, and signs through the local keystore daemon (SIGNING=execute),
falling back to compose mode when the daemon is not running.
Keep your project’s commands
Section titled “Keep your project’s commands”VibeKit does not replace a project’s build, test, or deployment scripts. Treat
its package.json as the command source of truth. Use VibeKit for agent setup,
LocalNet, keystore-backed signing, and on-chain operations.
To run without prompts in automation, pass explicit selections:
vibekit init --yes --agents claude --skills allUse vibekit --help for the complete headless flags: --agents <csv> (required
with --yes), --skills all|none|<csv>, --mcps none|<csv>, and --overwrite. Existing agent files
remain in place during headless setup unless you explicitly pass --overwrite.
Check the setup
Section titled “Check the setup”vibekit doctorStart with the read-only diagnosis. Use vibekit doctor --fix only when you
want it to repair the supported configuration it reports.