The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use shell commands (curl, file, and shell test operators) to manage icon files. These commands use variables such as <icon_url>, <slug>, and <upload_url> which are sourced from external web searches or third-party MCP tool responses. Without explicit sanitization, this pattern creates a risk of command injection if the agent interpolates malicious strings into the shell execution environment.
[EXTERNAL_DOWNLOADS]: The workflow relies on WebSearch and WebFetch to locate and download image assets from arbitrary third-party domains. While the skill suggests preferring official sources, it ultimately allows downloading from any URL returned by a search engine, exposing the environment to untrusted content.
[DATA_EXFILTRATION]: The skill utilizes curl --upload-file to transmit local files from /tmp/ to a remote <upload_url>. An attacker who can influence the upload_url via a search result or a compromised tool response could potentially exfiltrate data from the agent's accessible filesystem.
[PROMPT_INJECTION]: The skill processes untrusted data from web search results to identify icons for download and processing.
Ingestion points: External icon URLs and filenames derived from WebSearch results.
Boundary markers: None identified; search data is used directly in shell commands and tool parameters.
Capability inventory: File system access (write to /tmp/), network access (curl), and multi-modal image generation tools.
Sanitization: Limited to basic file type verification using the file command and checking for non-empty files.

youtube-thumbnail-maker