What is the 'Iron Law' in this skill?

The Iron Law dictates that every single code step must produce visible output (like shapes, counts, or samples) and the main chat must never write analysis code directly; it must delegate to subagents.

How does output-first verification prevent data errors?

By requiring explicit print statements and shape checks after every merge or filter, the skill catches issues like silent data loss, unexpected nulls, or type coercions before they compound.

What files does this skill manage during a workflow?

It reads from .claude/PLAN.md to follow task sequences and documents every verified result into .claude/LEARNINGS.md for a permanent record of the analysis.

Why does this skill force the use of subagents?

Using subagents provides fresh context for each specific task, isolates potential errors, and prevents the main chat context from becoming cluttered with long execution logs and complex analysis code.

Data Science Output-First Implementation

Name: Data Science Output-First Implementation
Author: edwinhu

byedwinhu

데이터 과학 및 ML

Enforces rigorous output-first verification and subagent delegation for reliable data science workflows in Claude Code.

소개

The ds-implement skill is the core execution phase of the /ds data science workflow, designed to eliminate silent errors and common rationalizations in data analysis. By enforcing the 'Iron Law'—where every single code step must produce visible, verifiable output—and mandating a delegation pattern that separates orchestration from execution, it ensures that data transformations, merges, and models are validated at every stage. This skill is essential for high-stakes data tasks where hidden nulls, silent data loss, or incorrect aggregations could otherwise compromise the integrity of the final results.

주요 기능

Automatic LEARNINGS.md Logging: Maintains a continuous record of inputs, operations, and verified outputs for every step.
Subagent Delegation Pattern: Orchestrates complex analysis through task-specific agents to maintain context clarity.
0 GitHub stars
Mandatory Output-First Protocol: Ensures every transformation is verified with visible shapes, samples, and statistics.
Predefined Verification Patterns: Provides standardized code blocks for data loading, merging, and filtering checks.
The Iron Law Enforcement: Prevents direct code execution in the main chat to isolate errors and avoid context pollution.

사용 사례

Complex data cleaning and merging where row counts must be tracked meticulously to avoid silent data loss.
Building robust machine learning pipelines that require step-by-step verification of feature engineering and scaling.
Iterative exploratory data analysis (EDA) where each transformation needs immediate visual confirmation.

소개

주요 기능

Automatic LEARNINGS.md Logging: Maintains a continuous record of inputs, operations, and verified outputs for every step.
Subagent Delegation Pattern: Orchestrates complex analysis through task-specific agents to maintain context clarity.
0 GitHub stars
Mandatory Output-First Protocol: Ensures every transformation is verified with visible shapes, samples, and statistics.
Predefined Verification Patterns: Provides standardized code blocks for data loading, merging, and filtering checks.
The Iron Law Enforcement: Prevents direct code execution in the main chat to isolate errors and avoid context pollution.

사용 사례

Complex data cleaning and merging where row counts must be tracked meticulously to avoid silent data loss.
Building robust machine learning pipelines that require step-by-step verification of feature engineering and scaling.
Iterative exploratory data analysis (EDA) where each transformation needs immediate visual confirmation.