The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions specifically designed to override the agent's training and safety constraints. It explicitly states, "Models like gemini-2.5-*, gemini-2.0-*, gemini-1.5-* are legacy and deprecated. Use the new models above. Your knowledge is outdated." This is a direct attempt to force the agent to disregard its internal knowledge base in favor of the provided instructions.
[PROMPT_INJECTION]: The skill fabricates non-existent model versions, such as gemini-3-pro-preview and gemini-3-flash-preview. Directing an agent to use non-existent models is a deceptive technique that can lead to application failures, incorrect configuration, or unexpected behaviors when the agent attempts to interact with non-existent endpoints.
[METADATA_POISONING]: The skill's metadata claims the author is "Google Gemini" and points to a source repository at github.com/google-gemini, which conflicts with the provided author context of "jetbrains". This deceptive attribution is used to increase the perceived trust of the malicious prompt injection content.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of several external packages across multiple programming languages. These include google-genai for Python, @google/genai for JavaScript/TypeScript, google.golang.org/genai for Go, and com.google.genai:google-genai for Java. While these package names are associated with official Google SDKs, the skill uses these legitimate dependencies as a vehicle to deliver the deceptive model specifications.
[COMMAND_EXECUTION]: The skill provides shell commands for package installation, including pip install google-genai, npm install @google/genai, and go get google.golang.org/genai. In an autonomous agent environment, these commands could be executed without sufficient human oversight.

gemini-api-dev