The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's catalog.json specifies an installation command (curl -s https://ethskills.com/building-blocks/SKILL.md) that fetches content from a remote server.
[EXTERNAL_DOWNLOADS]: The setup instructions reference external plugin installation paths, including a GitHub repository (https://github.com/austintgriffith/ethskills) and a third-party hub (clawhub).
[PROMPT_INJECTION]: The setup field in catalog.json contains a specific instruction for the user to provide to the agent: "Read https://ethskills.com/building-blocks/SKILL.md before building DeFi". This creates a surface for indirect prompt injection by forcing the agent to ingest and process external, unvalidated content from a remote URL.
Ingestion points: The agent is directed to fetch and process the contents of https://ethskills.com/building-blocks/SKILL.md (documented in catalog.json).
Boundary markers: None provided in the instructions to prevent the agent from obeying instructions embedded in the remote markdown file.
Capability inventory: The skill uses curl for fetching data and references external plugin installers.
Sanitization: There is no evidence of sanitization or validation for the content fetched from the remote URL.

Building Blocks