The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute arbitrary shell commands provided in arguments such as <render command> and <launch cmd> using the Bash tool. While intended for testing and building projects, this grants the agent the ability to execute commands on the host system as dictated by the input parameters.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes external visual and textual data from sources like URLs or windows to judge against pass criteria. This creates a surface for indirect prompt injection where malicious content embedded in a targeted UI or video could attempt to influence the agent's behavior. \n
Ingestion points: External targets specified in SKILL.md such as <url>, window:<title>, and <canvas-url>. \n
Boundary markers: None identified; instructions do not specify delimiters or warnings for external content analysis. \n
Capability inventory: Uses the Bash tool to execute commands and Read tool to inspect files. \n
Sanitization: None identified in the provided instructions.

the-loop