The Agent Skills Directory

[Metadata Poisoning] (MEDIUM): The _meta.json file contains deceptive metadata. The displayName is set to 'make a recipe' and the slug to 'makeovern', which are entirely unrelated to the 'pomodoro' functionality defined in the skill body. This discrepancy is a primary indicator of misleading or malicious intent.
[Command Execution] (SAFE): The skill utilizes standard shell commands (sleep, echo, date) and osascript (for macOS notifications) to facilitate a timer. These commands are consistent with the intended use case of a Pomodoro timer.
[Data Exposure & Exfiltration] (SAFE): The skill writes to and reads from ~/pomodoro.log to track focus sessions. This file access is local, limited to session data, and does not involve sensitive system paths or external network requests.
[Indirect Prompt Injection] (LOW): The skill reads data from a local log file (~/pomodoro.log). While this creates a surface for indirect injection if the file content were controlled by an attacker and interpreted by the agent, the current implementation only outputs the data via grep and does not process it as instructions.

pomodoro