pudl
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMPERSISTENCEDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill instructions direct the agent to offer to modify the user's shell startup files (such as
~/.zshrc,~/.bashrc, or~/.profile) to persistently set the$PUDL_DATAenvironment variable. - [DYNAMIC_EXECUTION]: The skill includes a local Python script
scripts/fetch_descriptor.pyand instructs the agent to execute it to manage metadata caches, involving both file system and network operations at runtime. - [EXTERNAL_DOWNLOADS]: The skill fetches data and metadata descriptors from vendor-controlled S3 buckets (
pudl.catalyst.coop) and official documentation sites. These are documented as vendor resources and do not independently escalate the verdict. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data sources, including Parquet files from S3 and documentation files (PDF, Word) converted to markdown. It lacks explicit sanitization steps or prompt boundary markers to mitigate risks from potentially malicious content embedded in these external datasets.
Audit Metadata