The Agent Skills Directory

[COMMAND_EXECUTION]: Unsafe shell variable expansion in curl commands. In the /deepl translate implementation, the variable $SOURCE_PARAM is used unquoted within the curl command line. Because this variable is constructed using the SOURCE_LANG input, an attacker could potentially perform argument injection to modify curl's behavior.
[COMMAND_EXECUTION]: Insecure file modification using sed. The skill updates XLIFF files by inserting translated text (sourced from the DeepL API) into a sed command using a pipe | delimiter. If a translation contains the delimiter character, it can break the sed syntax, potentially leading to file corruption or execution of unintended sed commands.
[REMOTE_CODE_EXECUTION]: Vulnerability to XML External Entity (XXE) attacks. The skill uses xmllint and Python's xml.etree.ElementTree to process user-provided XLIFF files. These libraries are typically vulnerable to XXE attacks if they are not specifically configured to disable external entity resolution, which could lead to local file disclosure or server-side request forgery (SSRF).
[DATA_EXFILTRATION]: Arbitrary local file access and upload. The /deepl file and /deepl xliff commands allow the agent to read any file from the local filesystem based on a user-provided path and upload its content to the DeepL API. This presents a risk of sensitive data exposure if used to target system files or private credentials.
[PROMPT_INJECTION]: Indirect prompt injection surface.
Ingestion points: User-provided text for translation and user-specified file paths for document translation (SKILL.md).
Boundary markers: No boundary markers or instructions to ignore embedded commands are present in the processing logic.
Capability inventory: The skill has the ability to read and write files (sed, cp, curl -o) and make network requests (curl).
Sanitization: There is no sanitization of the content returned from the translation API before it is written back to the filesystem or displayed.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to the DeepL API (api.deepl.com or api-free.deepl.com) to perform translation tasks.

deepl