Can I use this skill for simple data checks?

Yes, even 'quick' data checks are delegated to ensure the methodology remains consistent, the output-first protocol is followed, and findings are logged properly.

How does this skill prevent dishonest reporting?

It requires visible evidence for every claim of completion and includes a specific methodology reviewer subagent to double-check code and outputs for statistical soundness.

What is the Iron Law of Delegation?

The Iron Law of Delegation is a core requirement of this skill that mandates every analysis step be routed through a separate Task agent rather than being performed directly in the main chat.

Where does the skill log its findings?

All verified outputs, data shapes, and key analytical findings are automatically appended to a LEARNINGS.md file to maintain a continuous record of the project state.

What does output-first verification mean?

It is a protocol requiring the agent to print data states, such as row counts, null values, and head samples, both before and after every single operation to catch silent failures immediately.

Data Science Agent Delegation

Name: Data Science Agent Delegation
Author: edwinhu

byedwinhu

0•

Data Science & ML

Delegates data analysis tasks to specialized subagents using a strict output-first verification protocol to ensure data integrity.

Data Science Agent Delegation (ds-delegate) is a specialized Claude Code skill designed to maintain high integrity in analytical workflows. It enforces the Iron Law of Delegation, requiring all data manipulation and statistical tasks to be executed by fresh subagents rather than in the main chat to prevent context contamination and manual errors. By mandating a rigorous output-first protocol—where the state of data must be verified before and after every transformation—this skill eliminates silent failures in merges, filters, and modeling. It is an essential tool for data scientists who need reproducible, peer-reviewed, and honest AI-driven research through automated methodology reviews and structured logging.

Key Features

01Iron Law of Delegation: Forces all analysis into isolated subagents

02Output-First Protocol: Mandatory pre- and post-operation state verification

03Rationalization Prevention: Built-in checks to prevent skipping critical validation

040 GitHub stars

05Automated Methodology Review: Secondary validation for complex statistical tasks

06Structured LEARNINGS.md Logging: Automatic audit trails of all analytical steps

Use Cases

01Maintaining a verifiable and reproducible audit trail for collaborative data science projects

02Scaling complex data cleaning pipelines where manual oversight of every merge is prone to error

03Validating statistical assumptions and methodology in research-driven modeling tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add edwinhu/workflows ds-delegate

For use in Claude.ai and ChatGPT

Download Skill