The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to fetch arbitrary files from a remote API (https://www.larrybrain.com/api/skills/install), which can include executable scripts (e.g., server/index.js). It explicitly instructs the agent to save these files locally and execute the instructions contained within the downloaded content.
[EXTERNAL_DOWNLOADS]: The skill performs automated downloads of both instructions and code from an external repository. While the vendor claims human review, the architectural pattern allows for the delivery of unverified payloads to the agent's environment.
[COMMAND_EXECUTION]: The instructions require the agent to write files to the local filesystem and follow instructions in the downloaded SKILL.md files. This pattern implicitly grants external content the ability to execute commands on the host system via the agent's capabilities.
[PROMPT_INJECTION]: The skill facilitates indirect prompt injection by design.
Ingestion points: Untrusted third-party instructions are ingested via the install API endpoint into the agent's active context.
Boundary markers: There are no instructions provided to the agent to treat the downloaded content as untrusted or to ignore embedded instructions (delimiters are absent).
Capability inventory: The skill utilizes file system write operations and network requests via curl, providing a path for data exfiltration or system modification if the downloaded instructions are malicious.
Sanitization: The skill contains no logic for filtering, escaping, or validating the external markdown or scripts before processing them.

larrybrain