Every dependency an agent adds is code you now trust, and models sometimes invent package names that attackers then register, a tactic known as slopsquatting. Health signals like OpenSSF Scorecard and explicit rules for AI assistants help keep agent-chosen dependencies in check.
Read in this order
- PaperWe Have a Package for You! A Comprehensive Analysis of Package Hallucinations by Code Generating LLMsStart here: measured data on how often code models invent package names.
- OfficialSecurity-Focused Guide for AI Code Assistant Instructions (OpenSSF)Instruction rules you can give a coding assistant to pick safer dependencies.
- OfficialOpenSSF ScorecardAn automated health check you can run on any open source package.