Does this skill help with ML model calibration?

Yes, it specifically checks that plugins delegate probability math to reference logic (like Venn-Abers) to preserve the calibration guarantees of the Calibrated Explanations framework.

What is the main purpose of the ce-plugin-audit skill?

It is designed to ensure that plugins for the Calibrated Explanations library follow established architectural rules, ensuring valid probability math and metadata standards.

Which ADRs does this skill check?

The skill monitors compliance with several standards, including ADR-006 (metadata), ADR-015 (capabilities), ADR-013 (interval protocols), and ADR-001 (core boundaries).

Why does the skill check for lazy imports?

Lazy imports are checked to prevent heavy optional dependencies like Matplotlib or Pandas from increasing the package load time when those specific plugin features aren't being actively used.

Calibrated Explanations Plugin Auditor

Name: Calibrated Explanations Plugin Auditor
Author: Moffran

byMoffran

•

데이터 과학 및 ML

Audits and validates plugin implementations for the Calibrated Explanations framework to ensure compliance with architectural contracts and metadata standards.

The ce-plugin-audit skill provides a specialized auditing framework for developers working with Calibrated Explanations (CE). It automates the verification of plugin metadata, capability tags, and protocol adherence to ensure that extensions remain conformant with established Architecture Decision Records (ADRs). By checking for critical requirements like interval calibrator delegation, core-plugin boundary separation, and lazy dependency loading, this skill ensures that plugins maintain the rigorous calibration guarantees and performance standards required for reliable AI explainability.

주요 기능

0175 GitHub stars

02Architectural boundary checks to prevent leakage between core and plugin logic.

03Automated plugin_meta validation against ADR-006 and ADR-033 standards.

04Detection of silent fallbacks to ensure visible logging and warning compliance.

05Source code analysis for lazy loading of heavy dependencies like Matplotlib and Pandas.

06Protocol verification for classification and regression interval calibrators.

사용 사례

01Auditing third-party calibration plugins for trust rule compliance before registry inclusion.

02Verifying that a new explanation plugin correctly implements the required capability tags.

03Debugging non-conformant probability shapes in custom Venn-Abers implementations.

주요 기능

0175 GitHub stars

02Architectural boundary checks to prevent leakage between core and plugin logic.

03Automated plugin_meta validation against ADR-006 and ADR-033 standards.

04Detection of silent fallbacks to ensure visible logging and warning compliance.

05Source code analysis for lazy loading of heavy dependencies like Matplotlib and Pandas.

06Protocol verification for classification and regression interval calibrators.

사용 사례

01Auditing third-party calibration plugins for trust rule compliance before registry inclusion.

02Verifying that a new explanation plugin correctly implements the required capability tags.

03Debugging non-conformant probability shapes in custom Venn-Abers implementations.