What is the 'Iron Law' of ds-implement?

The Iron Law dictates that every single code step must produce visible output (prints, plots, or samples) to verify correctness before proceeding to the next task.

Why does this skill prefer output-first over TDD?

In data science, code can be syntactically correct but logically wrong (e.g., a merge dropping 50% of rows). Visual verification of data state is more effective than unit tests for catching these subtle issues.

How does it prevent silent data loss during merges?

The protocol mandates printing row counts and null counts immediately before and after every merge, ensuring any unexpected data drop is caught and documented instantly.

Where are the implementation insights stored?

All observations, output samples, and verification steps are automatically appended to a .claude/LEARNINGS.md file for full traceability of the project's evolution.

Data Science Implementation (Output-First)

Name: Data Science Implementation (Output-First)
Author: edwinhu

byedwinhu

0•

数据科学与机器学习

Enforces a strict output-first verification protocol for data science and machine learning implementation tasks.

ds-implement is the core execution engine of the data science workflow, designed to eliminate silent failures through a mandatory 'Output-First' protocol. Instead of relying on traditional unit tests, it requires every data operation—from loading and filtering to merging and modeling—to produce visible evidence such as data shapes, null counts, and sample distributions. By enforcing immediate verification and automated logging to a dedicated learnings file, it ensures that Claude maintains high-integrity data transformations and captures insights throughout the development process.

主要功能

01Structured Task Agent invocation for autonomous implementation steps

020 GitHub stars

03Real-time state monitoring for data shapes, dtypes, and null counts

04Automated documentation of transformations in LEARNINGS.md

05The Iron Law: Mandatory visible output for every code execution

06Standardized verification patterns for complex joins and merges

使用场景

01Building high-integrity data processing pipelines with verified transforms

02Developing machine learning models with transparent intermediate metrics

03Debugging complex data merges and aggregations where silent errors occur

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add edwinhu/workflows ds-implement

For use in Claude.ai and ChatGPT

Download Skill