The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes raw product reviews through its 'Local Review Toolkit'. 1. Ingestion points: Raw review text fetched from Amazon via the realtime/reviews endpoint in scripts/zoodata.py. 2. Boundary markers: The prompt rendering logic in scripts/zoodata.py includes explicit constraints for the model to only extract mentioned facts and respond in a strict JSON format. 3. Capability inventory: The skill is capable of performing API requests and data transformation via scripts/zoodata.py. 4. Sanitization: The model is instructed to translate non-English text and extract only specific semantic phrases, minimizing the risk of instructions in reviews being executed.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages authentication by retrieving ZOODATA_API_KEY from environment variables or designated configuration files such as ~/.zoodata/config.json. All network operations are directed to the official API at api.zoodata.ai, which is consistent with the skill's purpose and the vendor's infrastructure.

zoodata