help

WT Help

Display this reference card when invoked. One-shot, do not run wt setup, wt init, wt run, or edit config unless the user explicitly asks for execution.

Mental Model

Show this compact map in the user's language:

wt setup   = prepare this machine
wt init    = prepare this repo's wt state/config
wt run     = start work
autopilot = after ready, run, review, land, and settle the work
config  = inspect this repo and recommend the right config

Explain that wt run starts workspaces; it does not own review, landing, cleanup, or long-running coordination. Those are handled by autopilot and its phase skills.

Skills

Skill	Trigger	What it does
help	$help	This card.
config	$config	Inspect the current repo and recommend wt config.
ready	$ready	Turn intent into an approved launch target.
work	$work	Launch or coordinate prepared wt work.
land	$land	Review, land or discard, and settle lessons.
autopilot	$autopilot	Run the ready-approved path through work and land.
variants	$variants	Compare multiple implementation variants.

Setup

Machine/user concerns such as shell integration, agent hooks, cmux, and agent CLIs belong to setup and doctor:

wt setup
wt doctor
wt doctor --profile codex

Common Starts

wt run branch <words...>
wt run issue
wt run issue PROJ-123 --base .
wt run pr
wt run task
wt run workflow

Explain each command in one short phrase if helpful.

Deeper Model

Only when the user asks for more detail, explain these concepts compactly:

• TaskDocument: a local executable work unit stored under <repo-root>/.wt/execution/tasks/.
• Workflow: a saved execution plan that groups TaskDocuments or provider issues.
• TaskRun: the record of one launched execution.
• Modes: single shares one workspace, batch runs independent branches, stack runs ordered dependent branches, and matrix runs one task through multiple profiles.
• autopilot loop: $ready -> $work -> $land.

Keep the deeper model conceptual unless the user asks for concrete commands.

Guardrails

• Prefer "what owns what" over "everything wt can do".
• If the user is already inside a repo and asks what to do next, end with: "Run $config here first."
• If the user is ready to do real work, suggest $autopilot after config is healthy.