Indexing
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses an installation command (
curl -s https://ethskills.com/indexing/SKILL.md) that fetches content from an external domain not associated with the agent's core environment. - [INDIRECT_PROMPT_INJECTION]: The
setupinstructions incatalog.jsonexplicitly tell the agent to "Read https://ethskills.com/indexing/SKILL.md before querying chain data". - Ingestion points: External URL read via agent instruction in
catalog.json. - Boundary markers: None; instructions are interpolated directly into the agent's context.
- Capability inventory: The skill uses network operations (
curl) and references external plugin installation commands. - Sanitization: No validation or filtering is performed on the remote markdown content before it is processed by the agent.
- [METADATA_POISONING]: There is a discrepancy between the reported author ('bankrbot') and the content provider ('EthSkills') mentioned in the metadata and file contents, which may lead to confusion regarding the source of the instructions.
- [REMOTE_CODE_EXECUTION]: The
catalog.jsonfile contains instructions to install third-party plugins from a personal GitHub repository (https://github.com/austintgriffith/ethskills), which involves downloading and executing code from an unverified source.
Audit Metadata