The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the breathe.py script directly from a personal GitHub repository (marekkowalczyk/breathe-cli) that is not a recognized trusted vendor or well-known service.
[REMOTE_CODE_EXECUTION]: The installation process involves downloading a Python script and immediately executing it (curl -O ... && ./breathe.py), which constitutes the execution of remote code from an unverified source.
[COMMAND_EXECUTION]: The skill documentation describes using subprocess.run to call the breathe command-line tool within Python scripts. It also suggests integration with system tools like terminal-notifier and afplay via the command line.
[PERSISTENCE]: The documentation provides specific instructions for creating a crontab entry to schedule daily reminders, which is a standard method for establishing persistence on a system.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to how it handles session parameters.
Ingestion points: User-provided inputs for parameters like --duration and --ratio are interpolated into shell commands in the provided examples (SKILL.md).
Boundary markers: There are no boundary markers or instructions to ignore embedded commands within these input fields.
Capability inventory: The skill has the capability to execute shell commands, perform network downloads, and write to the file system for logging purposes.
Sanitization: There is no mention of input sanitization or validation of these arguments before they are passed to the shell.

breathe-cli-resonance-breathing