liquidium-borrow
Warn
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill's CLI wrapper (scripts/liquidium_cli.mjs) automatically performs a package installation using 'npm ci' if the core dependency is missing from the local cache. This involves downloading the @liquidium/client package and its dependency tree from the public npm registry at runtime.
- [DYNAMIC_EXECUTION]: The script uses 'createRequire' to dynamically load the @liquidium/client module from a computed path in the user's home directory (~/.cache/liquidium-borrow/node). This runtime loading of externally fetched code is a potential vector for executing unauthorized code if the cache directory is compromised.
- [INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting data from the Liquidium protocol API, such as pool descriptions, loan activities, and market states. This data is then presented to the agent, creating a surface where malicious content embedded in the API responses could attempt to override agent instructions or misguide user financial decisions. The skill mitigates this through explicit instructions requiring user confirmation for all transactions.
Audit Metadata