The Agent Skills Directory

[COMMAND_EXECUTION]: The template save command in scripts/publisher.sh is vulnerable to path traversal. The --name argument is used directly in a cp command without sanitization, enabling the creation or overwriting of files outside the designated templates/ directory (e.g., ../../scripts/malicious.sh).
[DATA_EXFILTRATION]: The scripts/post.sh script contains a vulnerability where metadata fields, such as 'title', are passed to curl -F without sanitizing for the '@' character. If a maliciously crafted title starting with '@' is imported and subsequently published, curl will interpret it as a file path, reading and uploading the contents of that local file to the external API.
[EXTERNAL_DOWNLOADS]: The skill uses curl to interact with an external API at https://api.upload-post.com/api. While this is core to its functionality, the lack of input validation on parameters makes this a potential vector for exfiltrating sensitive data to the remote server.
[PROMPT_INJECTION]: The skill's import logic in scripts/import-genviral.sh and scripts/import-dr-nora.sh presents an indirect prompt injection surface. It ingests untrusted data from external JSON files and directories without boundary markers or robust validation. This data is then used in powerful downstream tools like ffmpeg and curl without sufficient sanitization beyond basic character filtering.

social-publisher