mantle-openclaw-competition
Warn
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install a package directly from a third-party GitHub repository (
github:mantle-xyz/mantle-agent-scaffold) and execute it. This introduces an external dependency that is not verified or pinned to a specific version, presenting a potential supply chain risk. - [METADATA_POISONING]: There is a discrepancy between the author identified in the skill metadata (
mantle-xyz) and the author provided in the instruction context (starchild-ai-agent), which may cause confusion regarding the skill's origin and security standing. - [INDIRECT_PROMPT_INJECTION]: The skill processes data returned from external blockchain queries via the
mantle-clitool, creating an injection surface. \n - Ingestion points: Data returned from
mantle-cli catalog,mantle-cli swap pairs, andmantle-cli lp find-poolsas referenced inSKILL.md. \n - Boundary markers: The instructions mandate using
--jsonoutput for machine parseability but lack explicit directives for the agent to sanitize or ignore potentially malicious instructions embedded in blockchain metadata (e.g., token names or pool descriptions). \n - Capability inventory: The skill possesses significant capabilities, including performing on-chain financial transactions and interacting with various DeFi protocols. \n
- Sanitization: No validation or filtering mechanisms are specified for the external data ingested from the CLI tool before it is utilized to inform agent logic.
Audit Metadata