The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/upload-clipboard.sh performs unsafe path expansion using eval echo "$line" on strings obtained from the macOS clipboard via pbpaste. This vulnerability allows for arbitrary shell command execution if the clipboard content contains shell metacharacters like backticks, subshells, or semicolons.
[COMMAND_EXECUTION]: The script scripts/upload-clipboard.sh constructs a command string by appending unquoted file paths derived from the clipboard and executes the resulting string using eval $CMD. This provides a secondary vector for command injection.
[DYNAMIC_EXECUTION]: The Python script scripts/upload.py automatically installs the alibabacloud-oss-v2 package from the official repository using subprocess.check_call if it is not found in the local environment at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface by ingesting untrusted data from the system clipboard and standard input (stdin) and using it to define execution arguments.
Ingestion points: Clipboard content via pbpaste in scripts/upload-clipboard.sh; standard input in scripts/upload.py via sys.stdin.
Boundary markers: None implemented for untrusted inputs.
Capability inventory: The skill has capabilities for file system access, network operations (Alibaba Cloud OSS), and command execution through shell eval and Python subprocess.
Sanitization: There is no sanitization or escaping of the ingested data before it is passed to shell execution contexts.

alioss-upload