The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages the ZOODATA_API_KEY credential using standard environment variables or a local configuration file at ~/.zoodata/config.json. The included Python script (scripts/zoodata.py) implements a _is_trusted_host check that strictly limits API key transmission to vendor-owned zoodata.ai domains or localhost, preventing accidental exfiltration to third-party hosts.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted product reviews fetched from Amazon.
Ingestion points: Raw review data is ingested via the reviews-raw and realtime/product subcommands in scripts/zoodata.py.
Boundary markers: The script uses triple-quote delimiters (""") within the render_review_map_prompt function to isolate external review content from the agent's instructions.
Capability inventory: The skill uses network access to the vendor's API and subprocess execution for the bundled CLI.
Sanitization: External review text is truncated to 500 characters, reducing the attack surface for complex injection payloads.
[COMMAND_EXECUTION]: The skill invokes a local Python script (scripts/zoodata.py) to perform its data operations. The script is developed using only standard Python libraries and enforces a subcommand allowlist (scripts/allowed-commands.json), ensuring the agent only executes intended functions.

amazon-listing-audit-pro