The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided prompts and visual references to generate assets, which is a standard attack surface for indirect prompt injection. However, this is the intended core functionality and the risks are well-mitigated by design.
Ingestion points: External data enters the context via the prompt and reference parameters defined in SKILL.md.
Boundary markers: The skill requires the agent to call view_image to inspect local reference files before use, ensuring the data is visible to the system's safety filters.
Capability inventory: The skill's capabilities are restricted to image processing (using PIL and numpy) and writing assets/metadata to the project directory. It does not perform network operations or execute arbitrary shell commands.
Sanitization: The generate2dsprite.py script includes a sanitize_slug function that filters filenames to prevent path injection or filesystem issues.
[DYNAMIC_EXECUTION]: The provided Python scripts use json.loads for configuration and do not employ dangerous functions like eval(), exec(), or unsafe deserialization (e.g., pickle). All execution is deterministic and based entirely on the local code provided with the skill.

generate2dsprite