schema-induction

Compare labeled instances to classify fields as SAME or VARIES.

Updated Apr 20, 2026

Search Tags:#configuration #reverse engineering #data understanding #API response #validation contract #schema induction #pairwise contrast

Ready to run

npx skills add https://github.com/thistleknot/skills --skill schema-induction

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: schema-induction
Source: https://github.com/thistleknot/skills/tree/main/schema-induction
Command: npx skills add https://github.com/thistleknot/skills --skill schema-induction

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Schema-induction helps you discover the real structure of an undocumented format by comparing multiple examples and separating what is constant from what varies.

Core Features & Use Cases

Pairwise SAME/VARIES partitioning: Compares fields across every instance pair to identify invariants and differentia-driven variance.
Aggregated minimum schema: Produces a practical contract of required fields (constants + universal variables) plus optional conditional extensions.
Range and type characterization: Infers observed types and value ranges (or enumerations) to guide downstream validation and debugging.
Bug isolation across instances: Uses the schema-level differentia as a structured hypothesis space for fixing a regression.

Quick Start

Run schema-induction with at least 3 labeled instances of the same kind to generate the minimum schema and optional extensions you can then validate against new inputs.

Frequently Asked Questions about schema-induction

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I reverse engineer an undocumented API response schema?▼

To reverse engineer an undocumented API response, provide at least three labeled instances. The schema is induced by comparing fields across instance pairs to classify values as constant or variable, generating a minimum validation contract.

What is pairwise contrast for data understanding?▼

Pairwise contrast for data understanding compares fields across every instance pair to identify invariants. It separates constant fields from differentia-driven variables to discover the real structure of structured documents.

How do I generate a validation contract from multiple configuration files?▼

To generate a validation contract from multiple configuration files, aggregate constants and universal variables from provided examples. The process characterizes observed types and ranges to emit a minimum schema with optional conditional extensions.

How many examples do I need to induce an unknown structured schema?▼

You need at least three labeled examples to induce an unknown structured schema. This minimum sample size allows the pairwise partitioning loop to effectively separate invariants from variables and characterize value ranges.

Can I use schema induction to fix a regression across file artifacts?▼

Yes, schema induction helps fix a regression across file artifacts by using schema-level differentia as a structured hypothesis space. It isolates bugs by comparing multiple instances to identify what varies between working and broken outputs.