The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bundled Python CLI script (scripts/zoodata.py) to perform its primary functions. A maintenance routine, _schedule_stale_result_cleanup, uses subprocess.Popen to launch a detached background worker for deleting old temporary files. This execution is limited to the script's own maintenance, uses hardcoded paths and flags, and does not incorporate user-provided input, making it a low-risk internal operation.
[INDIRECT_PROMPT_INJECTION]: The skill processes Amazon product reviews retrieved from an external API, creating a potential surface for indirect prompt injection.
Ingestion points: Data is ingested from the api.zoodata.ai service via the reviews-raw command and processed using templates in review-tag-prompt.
Boundary markers: The skill uses explicit instructional delimiters (triple backticks) and strict JSON output requirements for the agent to mitigate the risk of the LLM following instructions embedded in the reviews.
Capability inventory: The agent has capabilities for local file reading/writing (for temp results and opted-in monitoring) and execution of the local CLI script.
Sanitization: The skill prioritizes structured JSON handling and uses standard libraries to sanitize inputs before they are incorporated into analysis prompts.
[SAFE]: Credential Handling and Host Validation. The script zoodata.py implements a trusted host check (_is_trusted_host) that ensures the ZOODATA_API_KEY is only transmitted to the vendor's official domain (zoodata.ai) or localhost. If an untrusted ZOODATA_BASE_URL is configured via environment variables, the script refuses to send the bearer token.
[SAFE]: Local File Management. The skill manages temporary API results in a private namespace within the system's temporary directory, using restrictive permissions (0700) to ensure data is only accessible to the current user. An automated garbage collection worker removes these files after 30 days.

amazon-market-analysis