The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes PowerShell commands using parameters derived from local configuration files (.v8-project.json) and user-supplied arguments (<ReportName>, SrcDir, OutDir). There are no instructions for the agent to sanitize or escape these inputs, creating a risk of command injection if an attacker provides shell metacharacters.
[REMOTE_CODE_EXECUTION]: The skill execution logic depends on a PowerShell script (epf-build.ps1) located in a sibling directory (../epf-build/scripts/). This reliance on external files outside the skill's own package boundary makes the behavior unverifiable and potentially dangerous if the sibling directory contains untrusted content.
[CREDENTIALS_UNSAFE]: The documentation and examples explicitly guide the user to pass database passwords (e.g., -Password "secret") as command-line arguments. This is an insecure practice as arguments are often logged in plain text in system process lists and shell history files.
[DATA_EXPOSURE]: The skill accesses and processes project-specific configuration files (.v8-project.json) and database connection details. While this is necessary for its function, the lack of input validation when resolving paths (like v8path) could lead to path traversal or execution of unintended binaries.

erf-build