The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the $ARGUMENTS variable in a shell command string (uv run python3 $CLAUDE_PROJECT_DIR/tools/scaffold.py $ARGUMENTS) without any quoting or sanitization. This allows for shell command injection if a user or a malicious prompt provides shell metacharacters such as ;, &&, |, or backticks. An attacker could execute arbitrary commands with the agent's privileges.
[REMOTE_CODE_EXECUTION]: The skill triggers the execution of a Python script (tools/scaffold.py) located in the project directory using dynamic input. Although the script is a vendor resource (associated with the author H-mmer), the method of execution allows for arbitrary code execution via the aforementioned shell injection vector.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to unvalidated interpolation of external data.
Ingestion points: The $ARGUMENTS variable in SKILL.md captures external input.
Boundary markers: No boundary markers or delimiters are used to isolate untrusted input.
Capability inventory: The skill allows shell command execution (uv run) and instructions the agent to perform multiple file-system operations (creating directories and placeholders).
Sanitization: There is no evidence of sanitization, validation, or escaping of the input arguments before they are used in the command.

new