The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources, specifically recommending the installation of the Bun runtime from the official bun.sh domain. Additionally, the Markdown processing script is capable of downloading remote images to a local temporary directory for inclusion in articles.
[REMOTE_CODE_EXECUTION]: Both the documentation and the environment check script suggest executing a remote installation script via a piped shell command (curl -fsSL https://bun.sh/install | bash). While targeting a well-known runtime provider, this involves the execution of remote code downloaded at runtime.
[COMMAND_EXECUTION]: The skill extensively uses subprocesses to launch Google Chrome with automation-related flags and to execute system utilities across macOS, Linux, and Windows. This includes the use of osascript for UI automation, swift for native clipboard access, and powershell.exe for Windows system calls to send keystrokes.
[DYNAMIC_EXECUTION]: For macOS compatibility, the skill dynamically generates Swift source code at runtime, writes it to temporary files, and executes it using the swift interpreter to perform native clipboard operations. This method is used to handle complex data types like images and rich text formatting that are not easily managed through standard command-line tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes external Markdown files to publish articles, creating a surface for indirect prompt injection.
Ingestion points: Markdown files provided as arguments to the article publishing script (scripts/x-article.ts).
Boundary markers: The skill does not implement explicit boundary markers or instructions to the agent to ignore potentially malicious content embedded within the processed Markdown.
Capability inventory: The skill possesses significant capabilities, including persistent browser sessions, filesystem access, and the ability to execute platform-specific system commands.
Sanitization: While the Markdown parser includes basic HTML escaping for code blocks and link attributes, it does not provide comprehensive protection against adversarial data designed to influence the agent's behavior during content processing.

tuzi-post-to-x