The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to interact with browsers and live terminal outputs, which exposes it to potential indirect prompt injection vulnerabilities when handling untrusted data from web or TUI sources.
Ingestion points: Dynamic content loaded from third-party web pages via agent-browser and active terminal interactions monitored via tctl in SKILL.md.
Boundary markers: Absent; there are no explicit constraints or delimiter boundaries specified to prevent text instructions within target applications from altering execution logic.
Capability inventory: Ability to interact with native desktop systems, execute commands via tctl, and type inputs into applications.
Sanitization: Absent; text outputs from terminal sessions and web pages are processed directly without input validation or escaping mechanisms.
[PRIVILEGE_ESCALATION]: The documentation references the use of sudo apt-get install commands to configure environment dependencies such as ffmpeg, cage, and wtype. These are standard prerequisite setup procedures for GUI and terminal automation and align with the skill's primary purpose.

droid-control