The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/upload-clipboard.sh utilizes eval on untrusted data sourced from the system clipboard.
Specifically, the line EXPANDED=$(eval echo "$line" 2>/dev/null) executes content directly from the clipboard to perform shell expansion, posing a high risk of arbitrary command injection.
The final upload command in the same script is also executed using eval $CMD, where $CMD includes variables derived from external input.
[REMOTE_CODE_EXECUTION]: The main Python script scripts/upload.py performs a dynamic installation of the alibabacloud-oss-v2 package using subprocess.check_call if the library is not found at runtime. While the package belongs to a well-known service, the pattern of runtime package installation is a security concern.
[DATA_EXFILTRATION]: The skill requires and handles high-privilege Alibaba Cloud credentials (OSS_ACCESS_KEY_ID and OSS_ACCESS_KEY_SECRET). While these are necessary for the skill's functionality, they represent sensitive data that could be targeted for exposure.
[INDIRECT_PROMPT_INJECTION]: The skill features multiple ingestion points for untrusted data, including the system clipboard, stdin, and interactive user prompts.
Ingestion points: scripts/upload-clipboard.sh (clipboard via osascript and pbpaste), scripts/upload.py (get_paths_from_input function via stdin/input).
Boundary markers: None identified for file path parsing.
Capability inventory: Subprocess calls in upload.py, eval and file existence checks in upload-clipboard.sh, and network operations via the Alibaba Cloud SDK.
Sanitization: The Python script attempts to sanitize filenames with MD5 hashing, but the shell script lacks robust sanitization for the paths it processes.

alioss-upload