robustness-audit

Audit code for latent runtime failure modes using a fixed defect-class taxonomy.

Updated Nov 8, 2025
One-click install
npx skills add https://github.com/berad217/human-training --skill robustness-audit
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: robustness-audit
Source: https://github.com/berad217/human-training/tree/main/skills-source/robustness-audit
Command: npx skills add https://github.com/berad217/human-training --skill robustness-audit

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill eliminates the risk of shipping code with latent runtime failures that pass all unit tests but break in production, including crashes, silent error swallowing, edge case mishandling, resource leaks, concurrency hazards, and security boundary holes that only appear under real-world usage conditions.

Core Features & Use Cases

  • Defect-class guided audit: Uses a fixed 6-category taxonomy to prime distinct mental models for finding bugs, avoiding the "find all bugs" overwhelm that leads to noise or missed issues.
  • Parallel subagent dispatch: Splits code into 3-5 natural surfaces and runs parallel reviews to cover more ground without context bloat.
  • False positive verification: Actively checks high-severity findings against actual source code to filter out agent hallucinations, with a dedicated "verified wrong" section to build trust in results.
  • Use cases: Pre-deployment hardening passes, FMEA-style code reviews, assessing unfamiliar or long-unmaintained codebases, investigating silent production failures where tests are all green, and finding bugs that can't be reproduced via live testing (no hardware, remote environment, etc.).

Quick Start

Request a robustness audit of your target codebase from your AI assistant, specifying any particular surfaces or defect classes you want prioritized if relevant.

Frequently Asked Questions about robustness-audit

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

FAQPage Schema
How do I find runtime bugs that pass all unit tests before deploying?

A robustness audit surfaces latent runtime failure modes like crashes, silent error swallowing, and resource leaks in code that passes unit tests. It dispatches parallel subagents across code surfaces and confidence-filters findings to exclude nits.

What is FMEA-style code review and when do I need it?

FMEA-style code review uses a fixed 6-category defect taxonomy to find edge case mishandling, concurrency hazards, and security boundary holes. You need it for pre-deployment hardening passes and assessing unfamiliar or long-unmaintained codebases.

How can I audit an unfamiliar codebase for silent production failures?

You audit an unfamiliar codebase by running a robustness audit that splits code into 3-5 natural surfaces and reviews them in parallel. It actively verifies high-severity claims against actual source code to minimize false positives.

Does code hardening help with concurrency hazards and resource leaks?

Code hardening via a robustness audit specifically targets concurrency hazards and resource leaks alongside crashes and security boundary holes. It applies a defect-class guided taxonomy to prime distinct mental models for finding these specific bugs.

How do I minimize false positives in automated code review?

To minimize false positives in automated code review, a robustness audit actively checks high-severity findings against actual source code and filters out agent hallucinations. It provides a dedicated verified wrong section to build trust in the reported results.

Can I investigate production crashes without reproducing them in a live environment?

You can investigate production crashes without live reproduction by running a robustness audit on the source code. It identifies error path misreporting and edge case mishandling that only appear under real-world conditions when direct testing is unavailable.