The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes external data from API responses and user-supplied seller artifacts (CSV/screenshots).
Ingestion points: scripts/zoodata.py CLI output and user-provided ABA-SQP/Amazon Ads data.
Boundary markers: The skill implements a robust 'Universal Gate' sequence and 'Shared Analysis Contract' to delimit and validate data transitions.
Capability inventory: The skill performs network operations via zoodata.py and manages local temporary files.
Sanitization: Data is validated through a 'Structured Field Identity Gate' and interpreted using strictly defined metric semantics (e.g., references/metrics-market-profile.md).- [COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/zoodata.py) to interface with the ZooData API.
The script includes a background cleanup mechanism that spawns a subprocess using sys.executable to remove stale temporary files. This operation is restricted to the script's own path and does not incorporate untrusted input.- [DATA_EXFILTRATION]: The skill transmits a user-provided ZOODATA_API_KEY to api.zoodata.ai.
The script implements a host validation check (_is_trusted_host) to ensure credentials are only transmitted to authorized vendor domains or localhost, preventing accidental leakage to third-party hosts.

amazon-keyword-traffic-analysis