The Agent Skills Directory

[COMMAND_EXECUTION]: The command template in SKILL.md interpolates user-supplied arguments (ObjectName, Lang, SrcDir) directly into a powershell.exe call using double quotes. This creates a surface for command injection if the inputs contain shell metacharacters (e.g., ;, &, |) that could terminate the intended command and start a new one.
[COMMAND_EXECUTION]: Both the PowerShell and Python scripts construct file paths using ObjectName and SrcDir without validating against path traversal. A user could provide values containing .. sequences or absolute paths to target files outside the intended project directory.
[DYNAMIC_EXECUTION]: The skill relies on external logic in scripts/add-help.ps1 and scripts/add-help.py to perform its operations, executing these scripts with runtime arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data which could be attacker-controlled.
Ingestion points: Reads .v8-project.json, Configuration.xml, ParentConfigurations.bin, and form metadata XML files in scripts/add-help.ps1 and scripts/add-help.py.
Boundary markers: None used; the scripts parse XML and JSON data directly to drive execution logic.
Capability inventory: File system writes, directory creation, and command execution via PowerShell.
Sanitization: Language codes are validated against a whitelist, but the contents of ingested XML/JSON files and other parameters are not sanitized against path traversal or structured data injection.
[EXTERNAL_DOWNLOADS]: The Python implementation scripts/add-help.py depends on the lxml package, which must be present in the environment.

help-add