What is 'theory drift' in software development?

Theory drift occurs when the original design intent, requirements, and invariants of a system diverge from the actual implementation over time as code is modified.

Does this skill rewrite my existing code comments?

No, EPT follows an append-only philosophy. It adds new patch entries to your comments and documents to preserve the history of decisions and changes.

What happens if a code change contradicts existing documentation?

EPT will explicitly surface a 'conflict' warning with pointers to the discrepancy, requiring the developer to either update the code or record a new design decision.

How does EPT provide evidence for code changes?

EPT requires every claim—such as a requirement or invariant—to cite specific evidence, which can include test names, code location references, or design document section IDs.

Evidence-Patched Theory (EPT)

Name: Evidence-Patched Theory (EPT)
Author: rinfz

byrinfz

•

学習とドキュメンテーション

Maintains system integrity by appending evidence-linked requirement and design patches directly to code and documentation.

Evidence-Patched Theory (EPT) is a specialized documentation and alignment skill designed to prevent 'theory drift'—the divergence between code implementation and its original design intent. By operating as a PR-driven agent, EPT generates structured, append-only patches for code comments and design documents, ensuring every requirement, invariant, and architectural decision is backed by verifiable evidence from tests or PRD slices. This skill is essential for teams managing complex, long-lived codebases where maintaining a clear rationale for implementation choices is critical for long-term stability and reviewability.

主な機能

01Conflict detection between documentation, code behavior, and tests

02Automated PRD and design document addenda generation

03Structured code comment patching for requirements and invariants

04Drift prevention via 'needs grounding' alerts for unverified changes

05Mandatory evidence linking to specific tests, code locations, or design IDs

061 GitHub stars

ユースケース

01Preserving architectural rationale and business logic during high-velocity refactoring

02Automating the creation of a chronological decision log for complex system invariants

03Enforcing strict alignment between product requirements (PRDs) and implementation

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add rinfz/garza evidence-patched-theory

For use in Claude.ai and ChatGPT

Download Skill