The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file defines a PowerShell command execution pattern that interpolates user-provided placeholders (<Name>, <Synonym>, and <SrcDir>) directly into a shell command string. This creates a high risk of command injection if a user provides inputs containing shell metacharacters such as semicolons, pipes, or quotes to escape the intended command context.
[PROMPT_INJECTION]: The skill instructs the agent to execute a shell command using parameters derived from user input without providing any sanitization logic or boundary markers. This allows an attacker to manipulate the agent into executing arbitrary system commands by providing malicious values for the Name or SrcDir parameters.
[COMMAND_EXECUTION]: Both scripts/init.ps1 and scripts/init.py use the SrcDir and Name parameters to construct filesystem paths using Join-Path or os.path.join without performing any validation or canonicalization. This enables path traversal attacks, where a user could specify a path like ../../ to create or overwrite files outside of the intended project directory.

epf-init