The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using the npx skills utility to search for, install, and update agent capabilities.
[EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of external software packages from GitHub and other remote sources via the npx skills add command. While the documentation mentions examples from established organizations like Vercel and Anthropic, the tool itself can target any arbitrary repository.
[REMOTE_CODE_EXECUTION]: The skill specifies the use of npx skills add <package> -g -y to install new functionality. The -y flag is a critical concern as it explicitly bypasses human confirmation prompts, allowing for the silent installation and potential execution of remote code. The -g flag indicates a global (user-level) installation, increasing the impact of a potentially malicious package.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by ingesting and processing untrusted data from the open skills ecosystem.
Ingestion points: Data enters the system through search results from npx skills find and the contents of packages installed via npx skills add (found in SKILL.md).
Boundary markers: There are no explicit instructions for the agent to treat search results or package metadata as untrusted data or to use delimiters to prevent instruction leakage.
Capability inventory: The skill possesses the capability to execute shell commands (npx), install software, and modify the agent's runtime environment.
Sanitization: While the instructions provide manual heuristics for quality verification (checking install counts and source reputation), there are no automated sanitization or validation mechanisms for the content being processed.

find-skills