The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill facilitates the download of third-party packages from inference-sh/skills via the npx skills add command. This registry is not included in the pre-verified list of trusted sources, posing a supply chain risk.
[REMOTE_CODE_EXECUTION] (MEDIUM): Installing skills from a remote registry leads to the execution of external code within the agent's environment. If a package in the registry were compromised, it could lead to remote code execution.
[COMMAND_EXECUTION] (LOW): The skill is explicitly granted permission to execute npx skills * in a Bash shell. This capability is used to modify the agent's software configuration and install new tools.
[PROMPT_INJECTION] (LOW): Vulnerable to indirect prompt injection through search results.
Ingestion points: Data retrieved from npx skills search and npx skills list from the inference.sh registry.
Boundary markers: None present; the agent treats registry metadata as instructions for skill selection and installation.
Capability inventory: Access to system commands via Bash(npx skills *) and the ability to download/execute new code.
Sanitization: No evidence of sanitization or validation of the registry's output before processing.

related-skill