alchemy-cost-tuning
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved from the blockchain, which could contain malicious instructions designed to influence the agent.
- Ingestion points: Untrusted data enters the system through
getTokenMetadataandgetNftMetadataBatchcalls inSKILL.md(Step 2), which fetch token and NFT metadata from the Alchemy API. - Boundary markers: The instructions do not define any delimiters or provide guidance to the agent to ignore or isolate instructions found within retrieved blockchain metadata.
- Capability inventory: The skill is configured with
Write,Edit, andBashaccess, presenting a risk of unauthorized actions if an indirect injection is successful. - Sanitization: There is no evidence of string validation, filtering, or escaping for data returned from external API calls before it is used in application logic.
Audit Metadata