The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external data in the form of Amazon product reviews, which creates a surface for indirect prompt injection. Attackers could theoretically embed malicious instructions in reviews to influence the agent's analysis.
Ingestion Point: Product reviews are retrieved from the ZooData API using the reviews-raw command in scripts/zoodata.py.
Boundary Markers: The skill utilizes prompt templates in scripts/zoodata.py (e.g., render_review_map_prompt) that explicitly instruct the LLM to constrain its output to JSON and extract only explicitly mentioned data.
Capability Inventory: The skill executes a bundled Python script and performs network operations to api.zoodata.ai.
Sanitization: The zoodata.py script truncates input review text to 500 characters, reducing the risk of processing large or complex malicious payloads.
[COMMAND_EXECUTION]: The skill uses a bundled script, scripts/zoodata.py, to manage data flow and API communication. This script is written using only the Python standard library and does not use dangerous functions like eval() or subprocess to execute untrusted code.
[EXTERNAL_DOWNLOADS]: The skill fetches market and product data from api.zoodata.ai. This domain is a verified vendor resource for the author serendipityoneinc. The implementation includes a security check, _is_trusted_host, which ensures the ZOODATA_API_KEY is only transmitted to trusted zoodata.ai domains or localhost, protecting the user's credentials from exfiltration to arbitrary hosts.

amazon-opportunity-discoverer