amazon-review-intelligence-extractor
Pass
Audited by Gen Agent Trust Hub on Jul 24, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its 'Insufficient Data Fallback' workflow.
- Ingestion points: Raw Amazon reviews are fetched using the
reviews-rawsubcommand (stored in$WORK/raw.json) and subsequently processed by the agent to extract sentiment and thematic tags. - Boundary markers: The review content is wrapped in triple quotes (
"""{text}""") within therender_review_map_prompttemplate to provide some structural separation. - Capability inventory: The agent can execute shell commands via a local Python script, perform network requests to the vendor's API, and write to the local file system.
- Sanitization: The raw review text is interpolated into the prompt without explicit sanitization or filtering of adversarial instructions that might be embedded in the customer feedback.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python utility (
scripts/zoodata.py) to handle API authentication, data retrieval, and local aggregation tasks. - [EXTERNAL_DOWNLOADS]: The Python script performs network operations to
api.zoodata.aito retrieve market data, product specifications, and customer reviews. This is a vendor-owned resource associated with the author 'SerendipityOneInc'.
Audit Metadata