The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands that unsafely interpolate variables from external data sources. For example, integration-patterns.md contains the instruction curl -o public/assets/meme-ref.png "<image_url>" which uses an image_url from thread.json. Similarly, character-pipeline.md specifies node scripts/build-character.mjs "<Full Name>" using a name from web search results. These patterns are vulnerable to command injection if the inputs contain shell metacharacters such as backticks or subshell markers.
[EXTERNAL_DOWNLOADS]: The skill workflow depends on downloading assets from several external domains. This includes public asset repositories like kenney.nl, opengameart.org, and itch.io, as well as arbitrary URLs found during web searches or provided in context files like thread.json.
[REMOTE_CODE_EXECUTION]: The combination of unsafe command interpolation and external data ingestion creates a surface for remote code execution. An attacker could provide a malicious string in thread.json or influence web search results to execute arbitrary commands on the system.
[COMMAND_EXECUTION]: Mandatory Evidence Chain for Indirect Prompt Injection surface: (1) Ingestion points: thread.json (image_url field) and web search results (personality photos). (2) Boundary markers: No delimiters or isolation instructions are present. (3) Capability inventory: The skill uses curl for networking and node to execute local scripts that interact with the file system. (4) Sanitization: No sanitization or validation of external input strings is specified before they are used in shell commands.

game-assets