allium
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines several shell-based workflows using
curlto interact with the Allium API atapi.allium.so, covering registration, polling for status, and data retrieval. - [INDIRECT_PROMPT_INJECTION]: The skill exposes a feature for running custom SQL against Allium's data warehouse via the
run-asyncendpoint. This allows for arbitrary query construction based on user parameters, which is a known vector for indirect prompt injection in data processing contexts. - [DATA_EXFILTRATION]: The skill's setup and operational instructions involve reading from and writing to
~/.allium/credentialsto manage API keys. This sensitive file access is part of the standard credential persistence mechanism for this tool. - [SAFE]: The skill explicitly defines handoff rules to direct users to alternative tools for requests outside of Allium's specialized scope, such as real-time pricing or NFT metadata.
Audit Metadata