tf-ralph

name	tf-ralph
description	Batch-process repetitive work with the Ralph pattern, tracking each item as a ticket.
allowed-tools	Bash, Read, Glob, Grep, Write, Edit

Marblo Ralph — Repetitive Work Automation

Handle repetitive work alongside Marblo tickets using the Ralph pattern.

---

⛔ Required rule: Marblo MCP only

Never use the Claude Code built-in tools (TaskCreate, TaskList, TaskUpdate, TaskGet). Task creation/lookup/status-change/logging MUST use the Marblo MCP tools:

• Bulk create: create_tasks_bulk (create all tickets in a single call)
• Work: claim_task, update_task_status, add_activity, submit_for_review
• Lookup: get_all_tasks, get_available_tasks

---

Usage

Ask the user for the following:

1. Targets: which files/components/endpoints to process repeatedly
2. Work: what to do to each target (add tests, refactor, improve a11y, etc.)
3. Project name: the project name to use on the Marblo tickets

Workflow

1. Analyze the target files/components and build a list.
2. Use create_tasks_bulk to create one ticket per target:
   • title: "[work type] target name" (e.g.: "[Test] /api/users")
   • role: the role matching the work
   • priority: set the same for all
3. Process the tickets one at a time, in order: a. claim_task → update_task_status(start_work) b. Do the actual work c. Record the result with add_activity (e.g.: "wrote 3 pytest tests, 3/3 pass") d. Success: submit_for_review → update_task_status(DONE) e. Failure: update_task_status(FAILED) + record the failure cause
4. After all tickets are processed, summarize:
   • DONE: N
   • FAILED: N
   • causes of failed items

Ralph vs Agent Teams

• Ralph: repeat the same work across N targets. 1 agent processes sequentially.
• Agent Teams: different work in parallel. N agents collaborate.