[run] — entrypoint

[run]
entrypoint   = "claude"
instructions = "cd {install_dir} && claude"

Fields

entrypoint (string)

The command users run to start the rig. For runtime = "claude-code" rigs this is typically just "claude".

This is mostly a string for documentation — rig does not currently spawn the entrypoint for you (running the agent is the user's job). It's surfaced in rig doctor output and the Hub package page so users know what to type.

instructions (string)

A human-readable startup instruction. Supports the placeholder {install_dir} which is replaced with the absolute path to the install directory at runtime.

Shown after a successful rig install:

Setup complete. Run:

  cd /Users/you/my-rig && claude

Why two fields

entrypoint is the program; instructions is the full sentence a human (or agent) reads. The latter is what gets shown to the user, but entrypoint is structured enough for the doctor JSON contract to surface programmatically.

Examples

Basic Claude Code rig:

[run]
entrypoint   = "claude"
instructions = "cd {install_dir} && claude"

Rig that wants you to source an env first:

[run]
entrypoint   = "claude"
instructions = "cd {install_dir} && source ./.env.local && claude"

A non-Claude rig (when more runtimes exist):

[run]
entrypoint   = "your-runtime --workspace {install_dir}"
instructions = "your-runtime --workspace {install_dir}"

What rig does not do

• Does not spawn the entrypoint as part of any command.
• Does not validate that entrypoint is runnable (use [tools] for that).
• Does not background-run the agent.

See also

• [tools] — declare claude as required.
• Quickstart: install