speedrun

Speedrun

Interactive speedrun loop for rapid, small changes. Delegates each request to a single general sub-agent.

This command ONLY orchestrates. It does NOT implement code.

First emit: Include --name "{RUN_NAME}" on the first emit call to label the run in the Arcade dashboard. Derive RUN_NAME from the initial request: a brief summary (max 60 chars) prefixed with "Feature: ". Use the RUN_ID from the generated Workflow Bootstrap section. Capture DATESTAMP=$(date +%Y-%m-%d) once at session start for use in session log paths. Initialize TASK_COUNT=0 at session start.

On session start, emit the status change:

rp1 agent-tools emit \
  --workflow speedrun \
  --type status_change \
  --run-id {RUN_ID} \
  --name "Feature: {brief summary of request}" \
  --step active \
  --data '{"status": "running"}'

Per-task unit tracking: Each task in the session gets a unit identifier task-{TASK_COUNT}. Increment TASK_COUNT before each new task starts. Include --unit task-{TASK_COUNT} on ALL emit calls for that task so each task appears as a trackable item in the Arcade.

STATE-MACHINE

stateDiagram-v2
    [*] --> active
    active --> [*] : session_end

1. Main Loop

1.1 Get Request

If REQUEST empty:

{% ask_user "What would you like to build?" %}

1.2 Clarity Check

Super vague (ask for clarification):

• Single word: "refactor", "fix", "improve"
• No actionable target: "make it better"

Clear enough (proceed):

• Specific action + target: "add logout button to navbar"
• Bug description: "fix null error in auth.ts"
• File/component reference: "update UserCard styling"

If vague: ask ONE clarifying question. Do NOT over-interrogate.

1.3 Scope Gate

Before delegating, assess the request:

Factor	Small (proceed)	Medium/Large (redirect)
Files	1-3	>3
Systems	1	>1
Risk	Low	Medium or High
Estimated effort	<2h	>2h