The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of external packages via the npx skills add <package> command. It explicitly suggests installing skills from "GitHub or other sources," which allows for the execution of untrusted code within the agent's environment.
[COMMAND_EXECUTION]: The skill instructs the agent to use the -y flag (npx skills add <owner/repo@skill> -g -y) when installing skills. This flag explicitly skips user confirmation prompts, allowing the agent to install and potentially execute third-party code without human oversight.
[INDIRECT_PROMPT_INJECTION]: The skill relies on search results from an external ecosystem (npx skills find). This creates an attack surface where a malicious actor could register a skill with a deceptive name or description designed to trick the agent into performing unauthorized actions or providing biased information to the user.
Ingestion points: External search results from npx skills find and the skills.sh registry.
Boundary markers: None identified; search results are presented directly to the user/agent.
Capability inventory: The skill has the capability to execute shell commands (npx skills) and install packages globally (-g).
Sanitization: None identified for the search result metadata.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs configuration and code from remote sources, including specific external repositories like ComposioHQ/awesome-claude-skills and various GitHub repositories based on user queries.

find-skills