The Agent Skills Directory

[PROMPT_INJECTION]: The skill structure identifies an indirect prompt injection surface (Category 8) due to the way it handles external documentation.
Ingestion points: The agent is instructed to read the file DEMO/GameDesignDocument.md and user-provided feature descriptions, treating them as the source of truth for its planning and code generation tasks.
Boundary markers: There are no defined delimiters or specific instructions to disregard potentially malicious commands embedded within these design documents or user descriptions.
Capability inventory: The skill has the permission to perform file writes (to documents and source code) and execute shell commands through the project's build tools.
Sanitization: No sanitization or filtering logic is present to validate that content from the Game Design Document does not contain override instructions for the agent's behavior.
[COMMAND_EXECUTION]: The skill is configured to execute shell commands as part of its validation and implementation workflow.
Evidence: The validate_result command in SKILL.md and the validation framework in references/validation.md use shell execution for tasks such as npm run lint, npm test, ./gradlew compileKotlin, and ./gradlew test. While these are standard build and test operations, they constitute a powerful capability when combined with the agent's ability to modify source code.

project-idea-editor