The Agent Skills Directory

[COMMAND_EXECUTION]: The command template in SKILL.md directly interpolates user-provided arguments into a shell command: powershell.exe ... -ObjectName "<ObjectName>" -FormName "<FormName>". This pattern is susceptible to command injection; an attacker could provide a value like MyObject" & calc & " to execute arbitrary system commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes external XML data which could serve as an attack vector for indirect prompt injection.
Ingestion points: The skill reads *.xml metadata and configuration files from the user-specified SrcDir (scripts/remove-form.ps1, scripts/remove-form.py).
Boundary markers: None identified in the prompt or data handling logic.
Capability inventory: The skill has the capability to recursively delete directories (shutil.rmtree, Remove-Item -Recurse) and overwrite files ([System.IO.File]::WriteAllText, open(path, "wb")).
Sanitization: There is no validation or sanitization of the content read from the XML files before it is used to determine which other files to modify or delete.
[DATA_EXPOSURE]: The skill implements recursive deletion logic based on user-supplied paths. Without strict path validation, a malicious or erroneous ObjectName or SrcDir could lead to the unintended deletion of sensitive project or system files.

form-remove