ralph-prd_skill

Overview

This skill generates structured prd.json files that break large work into independent, verifiable items following the Ralph Wiggum pattern. It produces a scoped instructions block plus per-item fields (id, category, description, file, steps, passes, skipped) so autonomous agents can execute, verify, and report progress. Use it to plan bulk migrations, refactors, or any batch task that benefits from explicit “done” criteria.

How this skill works

The skill inspects input sources such as file lists, lint output, or a user-defined scope and emits a prd.json with an instructions markdown and an items array. Each item is one independent unit (typically file-scoped) with concrete action and verification steps. Items start with passes:false and skipped:null so agents can update status as they complete or skip tasks.

When to use it

• Batch migrations: API upgrades, dependency bumps, codemods
• Large-scale refactors touching many files or components
• Lint or formatting campaigns that can be fixed per-file
• Test or type-fix sweeps where each file can be verified independently
• Any work you want tracked as discrete, verifiable items

Best practices

• Prefer per-file granularity to simplify verification and retries
• Embed clear before/after examples and constraints in instructions markdown
• Define precise verification commands (type-check, test, lint) for each item
• Include explicit skip conditions and reasons to avoid wasted attempts
• Group related items with a meaningful category to aid filtering and progress reports

Example use cases

• Generate PRD from eslint output to create one item per file with lint fix and type-check steps
• Create a migration PRD for a library upgrade, listing files that need import updates and verification commands
• Plan a refactor campaign by producing items per component with steps to update props and run related tests
• Prepare a test-stabilization PRD where each flaky test file becomes an item with reproduce and fix steps
• Bulk-rename or codemod work where each affected file is an item and verification runs the test or build