The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to automatically download and run code from external repositories using npx and uvx. Specifically, the command npx -y chrome-devtools-mcp@latest uses the -y flag to bypass the user confirmation prompt normally required for installing and executing packages, facilitating silent remote code execution.
[EXTERNAL_DOWNLOADS]: The skill triggers the download of multiple software packages from public registries (NPM and PyPI). It references external packages from non-trusted accounts such as cursortouch, stablyai, midscene-skills, and codex-skills, which are not verified vendors.
[COMMAND_EXECUTION]: The skill utilizes several system-level commands to perform its core functions, including npx, uvx, pip, and git. These are used to manage the environment and automate the installation of dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external Git repositories, which introduces a vulnerability to indirect prompt injection if the remote repository content or commit logs are manipulated by an attacker. \n
Ingestion points: Remote Git repository commit logs and metadata as described in the Guarded Update section of SKILL.md. \n
Boundary markers: None present to distinguish between instructions and remote data. \n
Capability inventory: The skill possesses network access, package installation capabilities, and command execution rights. \n
Sanitization: No sanitization or validation of the remote repository content is implemented; the skill relies on manual user verification of breaking changes.

agent-update-manager