The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The templates cloudflare-worker-template.sh, python-api-template.sh, and react-component-template.sh accept user-controlled input (Project Name) as a positional argument. This variable is used directly in shell commands.
Evidence: In templates/cloudflare-worker-template.sh, the variable $PROJECT_NAME is used in sed -i '' "s/PROJECT_NAME/$PROJECT_NAME/g" package.json. A malicious name containing shell metacharacters could lead to unintended command execution if the agent runs these scripts in a shell environment.
INDIRECT_PROMPT_INJECTION (LOW): The skill demonstrates a vulnerability surface where untrusted user input (the project name) is interpolated into scripts executed by the Bash tool.
Ingestion points: Positional arguments in templates/cloudflare-worker-template.sh, templates/python-api-template.sh, and templates/react-component-template.sh.
Boundary markers: Absent. No validation or sanitization of the $1 argument is performed before use in shell operations.
Capability inventory: Bash, Write, Read tools; scripts use mkdir, cat, and sed.
Sanitization: Absent. There is no escaping or validation logic for the input variable.
EXTERNAL_DOWNLOADS (SAFE): The scaffolded projects include standard dependencies from npm and PyPI. While these involve external network operations during setup, they are consistent with the primary purpose of a project generator and use well-known packages.

grey-haven-project-scaffolding