What problem does it solve?
This Skill solves the risk of security vulnerabilities becoming urgent CVE liabilities while preserving production stability, especially in GPU-heavy Python stacks like vLLM/PyTorch where dependency changes can easily break compatibility.
Core Features & Use Cases
- Four-step safe workflow: scan with pip-audit, classify upgrade risk, apply patch-bounded upgrades, then verify with targeted test tiers.
- Strict compatibility guardrails: prevents major/minor jumps in large dependencies (vLLM/torch and related GPU packages) and uses patch-only constraints like
<next-minor.
- Automation support: provides a script-driven approach to extract patch-only upgrade candidates from pip-audit JSON and optionally apply them with a re-audit check.
Example use case: you receive a Dependabot/Renovate alert for an aiohttp or cryptography CVE, and you need to remediate it quickly while ensuring the vLLM runtime and CUDA environment remain unchanged.
Quick Start
Run the skill command to apply patch-only fixes after your pip-audit scan: ./install.sh dependency-vulnerability-fix