The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads Node.js binaries and verification files from official nodejs.org servers and utilizes the Aliyun npm mirror (registry.npmmirror.com). These are established and well-known service providers.
[REMOTE_CODE_EXECUTION]: Installs the @iqinghu/qhkit package from the npm registry and executes Node.js binaries downloaded during the setup process. These actions target official and vendor-specific resources.
[COMMAND_EXECUTION]: Employs shell commands such as curl, tar, npm, and npx for environment bootstrapping, tool configuration, and image processing.
[DYNAMIC_EXECUTION]: Uses python -c and npx to generate and execute code at runtime for image resizing. This involves concatenating shell strings with file path variables, which is a high-risk pattern for command injection.
[INDIRECT_PROMPT_INJECTION]: The image compression instructions create a vulnerability surface for command injection. The agent is instructed to interpolate file paths (derived from user-provided images) directly into executable shell commands without any sanitization or the use of boundary markers.
Ingestion points: uploadedImages file paths provided by the user or external sources.
Boundary markers: None specified for the construction of shell command strings.
Capability inventory: python, npx, qhkit, and general shell access.
Sanitization: The skill lacks instructions for validating or escaping filenames before they are used as parameters in command execution.

linkpix-seedream-5-lite