The Agent Skills Directory

[COMMAND_EXECUTION]: Use of osascript for programmatic control of the Microsoft Excel UI on macOS. This capability allows the agent to manipulate the application window and active sheets via AppleScript instructions provided in the skill documentation.
[COMMAND_EXECUTION]: Execution of the system file utility via subprocess.run in scripts/parse_complex_excel.py. This is used to determine the MIME type of files before processing, representing a vector for system command execution.
[COMMAND_EXECUTION]: Vulnerable ZIP extraction in scripts/parse_complex_excel.py. The fix_defined_names function utilizes zf.extractall() on user-provided Excel files without performing path validation. This is susceptible to 'ZipSlip' attacks, where a malicious archive could overwrite sensitive files outside the intended temporary directory.
[PROMPT_INJECTION]: Deceptive metadata poisoning via the .security-scan-passed file. This file contains a fabricated security verdict and content hash, which is a common obfuscation tactic used to provide a false sense of security and bypass manual or automated review processes.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface (Category 8):
Ingestion points: scripts/parse_complex_excel.py reads cell data and XML structures from external .xlsx and .xlsm files.
Boundary markers: Absent. The skill does not use delimiters or instructions to differentiate between data and commands when processing Excel content.
Capability inventory: UI automation via AppleScript, file system writes via ZIP repackaging, and subprocess execution.
Sanitization: Absent. Extracted cell values and XML text are processed as raw strings without validation or sanitization before being displayed or used.

excel-automation