The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content and interpolates it into prompts for image generation, creating a vulnerability surface for indirect instructions.
Ingestion points: User-supplied captions (t1 through t9) and metadata provided in brand/ip-card.md are directly inserted into the image generation templates.
Boundary markers: No delimiters or instructions to ignore embedded commands are present in the prompt templates.
Capability inventory: The skill uses scripts/generate.py for network operations and scripts/slice_grid.py for file system writes to the out/ directory.
Sanitization: No evidence of input validation, escaping, or filtering for the interpolated user content.
[COMMAND_EXECUTION]: The skill requires the agent to execute local Python scripts (scripts/generate.py and scripts/slice_grid.py) to perform its primary tasks of image generation and image processing.
[EXTERNAL_DOWNLOADS]: The script scripts/generate.py performs network requests to fetch generated images. While it defaults to OpenAI's API, the skill documentation suggests an alternative third-party gateway (tryallapi.com) which is not a whitelisted or well-known trusted service.
[DATA_EXFILTRATION]: The skill transmits the sensitive GPTIMAGE_API_KEY to the configured GPTIMAGE_BASE_URL. The suggestion to use a non-whitelisted third-party service (tryallapi.com) as an endpoint for these requests introduces a risk of credential exposure to that service.

sticker-pack