The Agent Skills Directory

[PROMPT_INJECTION]: The instructions designate the skill as 'Self-Evolving', directing the agent to modify the skill's own file contents if logic or paths break. This encourages autonomous modification of instructions.
[COMMAND_EXECUTION]: The skill uses the open command to execute an application bundle (FloatingClock.app). The script attempts to launch from /Applications/ first, falling back to a local path within the user's home directory if the installed version is missing.
[PROMPT_INJECTION]: The instructions ask the agent to 'document the bypass' if a macOS Gatekeeper security prompt occurs during execution, which encourages the recording of security circumvention events.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to environment manipulation as it executes code based on the existence of specific file paths.
Ingestion points: File system directory existence checks in SKILL.md script.
Boundary markers: None provided to distinguish between valid and malicious app bundles at those paths.
Capability inventory: Access to the Bash tool to run open and echo commands.
Sanitization: No integrity checks (e.g., code signing verification) are performed on the application before execution.

launch