Stability Auditor

Audit numerical stability and nondeterminism in AI experiments and data pipelines.

1|Updated Feb 8, 2026
One-click install
npx skills add https://github.com/soheunyi/get-research-done --skill stability-auditor
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: Stability Auditor
Source: https://github.com/soheunyi/get-research-done/tree/main/skills/grd-stability-auditor
Command: npx skills add https://github.com/soheunyi/get-research-done --skill stability-auditor

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Audit numerical stability and randomness controls to keep outcomes reproducible and trustworthy. Use when results may be affected by seeds, nondeterminism, tolerances, or numerical precision.

Core Features & Use Cases

  • Detect nondeterminism and seed sensitivity in AI experiments.
  • Validate numerical stability across runs and under perturbations.
  • Use Case: Debug flaky experiments and ensure repeatable conclusions in simulations and data pipelines.

Quick Start

Run a stability audit on your latest experiment to detect nondeterminism and seed sensitivity.

Frequently Asked Questions about Stability Auditor

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

FAQPage Schema
How do I detect nondeterminism and seed sensitivity in AI experiments?

To detect nondeterminism and seed sensitivity, run a stability audit that validates numerical stability across runs and identifies whether your AI experiment outcomes shift based on random seeds or tolerances.

Why does my simulation produce different results across runs with the same inputs?

Your simulation produces different results due to numerical fragility and nondeterminism. A stability audit identifies hidden seed sensitivity and precision issues, applying deterministic settings and RNG metadata capture to fix the inconsistency.

What is numerical stability auditing and when do I need it for data pipelines?

Numerical stability auditing identifies nondeterminism and numerical fragility that can invalidate conclusions. You need it for data pipelines when results are sensitive to seeds, tolerances, or numeric precision across runs.

How do I ensure deterministic settings and reproducible results in AI experiments?

To ensure deterministic settings and reproducible results, apply a stability audit that captures RNG metadata and validates numerical stability under perturbations, guaranteeing repeatable conclusions across your AI experiments.

Can I validate numerical fragility under perturbations to debug flaky experiments?

Yes, you can validate numerical fragility under perturbations to debug flaky experiments. A stability audit applies perturbation-based validation to detect numerical instability and ensure repeatable conclusions in simulations.