The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The bash script in SKILL.md utilizes a broad file system search (find ~) to locate its execution script. This pattern is risky as it could identify and execute a malicious file with the same name if placed in the user's home directory.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill relies on data from a non-trusted GitHub repository (laris-co/nat-location-data). Since this source is external and unverified, the data (CSV files) could be manipulated to provide false information or embed malicious payloads.
[REMOTE_CODE_EXECUTION] (LOW): location-query.ts uses Bun's shell wrapper ($) to call the GitHub CLI (gh). While the current implementation uses hardcoded filenames, the use of a shell wrapper for API calls is a brittle pattern that can lead to command injection if improperly extended.
[PROMPT_INJECTION] (LOW): (Category 8) The skill creates a surface for Indirect Prompt Injection by processing untrusted CSV data without sanitization or boundary markers.
Ingestion points: location-query.ts fetches CSV content from GitHub.
Boundary markers: None. The data is parsed and directly interpolated into the output template in SKILL.md.
Capability inventory: The skill can execute shell commands via Bun and the GitHub CLI.
Sanitization: None. The manual CSV parser splits columns but does not validate or escape the content of the fields (e.g., 'address' or 'locality').

physical