The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill is highly vulnerable to Indirect Prompt Injection.
Ingestion points: User input from "Questions" and "Statements" (e.g., seed, hints parameters) as defined in SKILL.md and CARD.yml methods.
Boundary markers: None identified. User input is used directly to define the properties of generated rooms.
Capability inventory: Uses write_file and read_file to create and update directory structures and ROOM.yml files.
Sanitization: No sanitization or validation logic is present to prevent user input from containing malicious instructions for the agent.
[COMMAND_EXECUTION] (HIGH): The skill encourages unsafe file system operations. The 'Tower Directory Pattern' in SKILL.md explicitly utilizes directory traversal (e.g., ../../../../../skills/) to link floors to other skill directories. This capability allows the skill to potentially modify or read the configuration of other skills installed in the system, breaking intended isolation.
[DATA_EXFILTRATION] (MEDIUM): The use of read_file combined with the demonstrated directory traversal capability allows for the exposure of sensitive configurations from other skills.

world-generation