The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill calculates the file path for an executable script at runtime using a Python glob pattern.
Evidence: The command VM_PROVISION="$(python3 -c "import importlib.util, pathlib; p = pathlib.Path.home() / '.claude' / 'plugins'; print(list(p.glob('*/cloud-weaver/*/skills/cloud-weaver-vm-setup/scripts/vm-provision.py'))[0])")" followed by python3 "$VM_PROVISION" in SKILL.md dynamically locates and executes a script from the local file system.
[COMMAND_EXECUTION]: The skill performs several file system operations and remote connections via the shell.
Evidence: Commands like rm -f ~/.ssh/cloud-weaver*, ssh-keygen, and ssh -i ... ubuntu@"$public_ip" are used to manage local keys and remote VM access.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to command injection through unsanitized variable interpolation in shell scripts.
Ingestion points: The variables env_name and public_ip are used as inputs for shell commands.
Boundary markers: None identified; the instructions do not include delimiters or warnings to ignore embedded content in these variables.
Capability inventory: The skill has the capability to write/delete files (rm), execute local scripts (python3), and perform network operations (ssh).
Sanitization: There is no evidence of validation or escaping for the ${env_name} or "$public_ip" variables before they are interpolated into commands like rm -f ~/.cloud-weaver-"${env_name}"-*.json or ssh ... ubuntu@"$public_ip".

cloud-weaver-offboard