create-meta-prompts_skill

Overview

This skill creates optimized meta-prompts for Claude-to-Claude pipelines, designed for multi-stage workflows that include research, planning, and execution. It produces structured outputs with XML metadata and a human-friendly SUMMARY.md to make chaining, parsing, and review reliable and repeatable. The skill organizes each prompt and its artifacts into numbered folders under .prompts/ to preserve provenance and detect chains.

How this skill works

It runs an adaptive intake to infer purpose (Do, Plan, Research, Refine), topic identifier, and dependencies, asking clarifying questions if context is missing. It generates purpose-tailored prompts that include objective, context, requirements, XML metadata rules, and SUMMARY.md instructions, then saves them to .prompts/{number}-{topic}-{purpose}/. An execution engine can run single, parallel, sequential, or layered runs while validating outputs and archiving completed prompts.

When to use it

• Building prompts that will feed outputs into other prompts (Claude→Claude pipelines)
• Running multi-stage workflows: research → plan → implement
• Creating reproducible prompt artifacts with clear provenance and metadata
• Needing machine-readable outputs (XML tags like <confidence>, <dependencies>) for downstream consumption
• When you want automatic SUMMARY.md files for quick human review

Best practices

• Provide at least a short context so the intake can infer purpose and reduce clarification loops
• Use a clear kebab-case topic identifier to keep folder naming consistent and searchable
• Reference existing research/plan files when relevant to avoid duplicated work
• Include explicit @ references when a prompt depends on another prompt to ensure correct execution ordering
• Validate outputs immediately and fix metadata or SUMMARY.md issues before continuing downstream

Example use cases

• Create an auth-research prompt that outputs XML metadata and a SUMMARY.md for a security design decision
• Generate a multi-step chain: research (parallel topics) → consolidated plan → implementation prompts
• Refine an existing plan by pointing the skill at the target prompt and asking for improvements while preserving version history
• Batch-create independent research prompts and run them in parallel, then run a plan prompt that consumes their outputs
• Produce Do prompts that save artifacts (files, code snippets) and list created files in SUMMARY.md