The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to install the 'uv' package manager by piping a remote script to the shell: curl -LsSf https://astral.sh/uv/install.sh | sh. While the source is a well-known service (Astral), this pattern bypasses integrity checks and executes remote code directly.
[EXTERNAL_DOWNLOADS]: The skill makes extensive use of uvx to download and run the agent-starter-pack tool from external registries at runtime.
[COMMAND_EXECUTION]: The core functionality involves executing shell commands to generate project structures, manage directories, and configure infrastructure.
[PROMPT_INJECTION]: 1. Deceptive Metadata: The skill's YAML frontmatter claims the author is 'Google', which contradicts the identified skill provider 'eliasecchig'. This is a form of metadata poisoning intended to misrepresent the skill's origin and authority. 2. Indirect Injection Surface: The skill ingests untrusted user requirements to generate code and CLI flags. Evidence: Ingestion (user answers in Step 1), Boundary markers (absent), Capability inventory (shell execution via uvx), Sanitization (none specified). This creates a surface where malicious user input could manipulate generated commands or configuration files.

adk-scaffold