The Agent Skills Directory

[Dynamic Execution] (MEDIUM): The mcp__plugin_headless-knight_runCLI__gemini tool accepts an env object allowing for the definition of arbitrary environment variables. This capability can be exploited to hijack execution flow or escalate privileges using variables such as LD_PRELOAD, PATH, or NODE_OPTIONS within the execution context.
[Command Execution] (MEDIUM): The skill relies on an external environment variable GEMINI_CLI_COMMAND to determine the executable path. This creates a dependency on a computed path that could be manipulated to run malicious binaries if the environment is compromised.
[Indirect Prompt Injection] (LOW): The skill is explicitly designed to read and summarize external web pages and search results. This untrusted data can contain adversarial instructions intended to subvert the agent's logic.
Ingestion points: Web content and search results processed by the runCLI__gemini tool.
Boundary markers: No specific delimiters or safety instructions are defined to separate user/web data from the tool's system instructions.
Capability inventory: The skill possesses CLI execution capabilities and network access.
Sanitization: No sanitization or filtering logic is present for the ingested external strings.

run-gemini