Can I use this to improve existing plans?

Yes, the 'Refine' purpose is specifically designed to target existing research or planning files to iterate on and improve them based on new context or feedback.

How does this skill handle task dependencies?

It automatically scans generated prompts for file references and builds a dependency graph to determine the correct order for sequential or parallel execution.

What is a Claude-to-Claude pipeline?

A Claude-to-Claude pipeline is a workflow where the output of one Claude prompt serves as the structured input for another, allowing for complex multi-step reasoning and execution.

Do I need to read the full output of every research stage?

No, the skill generates a SUMMARY.md for every prompt execution, providing a concise 'executive summary' for quick human review while keeping full details in XML for Claude.

Where are the generated prompts and results stored?

All artifacts are organized within a local .prompts/ directory, using kebab-case naming and numeric prefixes to maintain a clear history of your project's evolution.

Meta-Prompt Workflow Architect

Name: Meta-Prompt Workflow Architect
Author: flosrn

byflosrn

0•

生産性とワークフロー

Orchestrates multi-stage Claude-to-Claude pipelines with automated research, planning, and structured execution stages.

The Meta-Prompt Workflow Architect is a specialized skill designed to manage complex, multi-step AI operations by generating optimized prompts for Claude-to-Claude communication. It breaks down large tasks into structured Research, Plan, and Implement stages, organizing them within a dedicated .prompts/ directory for perfect provenance. By utilizing XML metadata for machine parsing and generating SUMMARY.md files for human oversight, it enables developers to build sophisticated autonomous workflows without losing track of the chain of logic or intermediate findings.

主な機能

01Structured directory management with versioning and provenance tracking

02Intelligent 'Intake Gate' for adaptive requirement gathering and chain detection

030 GitHub stars

04Dependency-aware execution engine supporting sequential and parallel tasking

05Automated multi-stage workflow generation (Research, Plan, Implement, Refine)

06Automatic SUMMARY.md generation for rapid human review of AI outputs

ユースケース

01Managing iterative refinement cycles for AI-generated plans and implementations

02Creating automated documentation or roadmap strategies for legacy codebases

03Building complex software features that require deep architectural research before coding

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add flosrn/.claude create-meta-prompts

For use in Claude.ai and ChatGPT