The Agent Skills Directory

PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection due to the way it handles user input in brainstorming prompts.\n
Ingestion points: SKILL.md uses numerous placeholders (e.g., [topic], [your goal], [problem], [idea 1]) to insert untrusted data into instructions for the Gemini model.\n
Boundary markers: Absent. No delimiters or system-level instructions are used to distinguish user data from the core brainstorming rules.\n
Capability inventory: The skill uses the gemini CLI to perform network-based AI operations.\n
Sanitization: Absent. User input is processed as raw text within command-line strings.\n- EXTERNAL_DOWNLOADS (SAFE): The skill requires the installation of the google-generativeai package.\n
Evidence: pip install google-generativeai in SKILL.md.\n
Trusted Source: The package is maintained by Google, which is a trusted organization. Per [TRUST-SCOPE-RULE], this finding is considered safe.\n- COMMAND_EXECUTION (SAFE): The skill executes shell commands to interact with the Gemini CLI.\n
Evidence: Multiple code blocks in SKILL.md using the gemini command.\n
Context: This is the intended primary behavior of the skill and uses a trusted tool to perform legitimate brainstorming tasks.

brainstorm