The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: No patterns for remote script execution, such as curl | bash or eval() on remote content, were found. The skill relies on local code and standard library imports.
[DATA_EXFILTRATION]: No evidence of sensitive file access (e.g., .env, .ssh) or unauthorized network operations was detected. The mentioned network-related components (like PartyKit) are cited as related skills rather than being implemented as exfiltration vectors.
[PROMPT_INJECTION]: The instructions are strictly limited to game development logic and engine integration. There are no attempts to override agent behavior, bypass safety filters, or extract system prompts.
[OBFUSCATION]: The content is presented in clear, readable Markdown and TypeScript. No Base64 encoding, zero-width characters, homoglyphs, or other obfuscation techniques were identified.
[INDIRECT_PROMPT_INJECTION]: While the game handles user input via mouse and keyboard events, these are used exclusively for game mechanics (e.g., placing towers, menu navigation). The skill lacks the dangerous capabilities (such as arbitrary file writes or network sends) necessary to exploit this attack surface.

tower-defense-3d