The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill points to and facilitates downloads from an untrusted third-party GitHub repository: https://github.com/content-designer/ux-writing-skill.
[COMMAND_EXECUTION]: The skill provides explicit shell commands for the agent to run, including npx skills add and scope detection logic using bracketed shell tests.
The provided npx commands use the -y flag, which suppresses user confirmation prompts during the installation of external content.
The skill documentation encourages the use of the ! prefix (specific to certain agent environments) to execute shell commands directly within the conversation context.
[REMOTE_CODE_EXECUTION]: By invoking npx skills add on an untrusted repository, the skill enables the download and potential execution of arbitrary code or scripts on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process interface copy (such as error messages and notifications), which are often sourced from external application data, creating a potential surface for indirect prompt injection.
Ingestion points: UI copy elements (buttons, labels, error messages) provided as input for writing tasks (SKILL.md).
Boundary markers: None; the skill lacks delimiters or specific instructions to ignore embedded commands within the processed copy.
Capability inventory: Shell command execution via npx and the platform-specific ! syntax (SKILL.md).
Sanitization: There are no mechanisms described for sanitizing or validating the copy being processed for writing improvements.

ux-writing-skill