The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/crawlora.sh executes curl to interact with the Crawlora API. The script includes robust input validation: it restricts the character set of the CRAWLORA_API_KEY, uses a hardcoded base URL to prevent redirection, and strictly validates the requested path against a whitelist of allowed endpoints. It also uses --data-urlencode to prevent argument injection in the query string.
[CREDENTIALS_UNSAFE]: The skill follows security best practices for credential management. It requires the CRAWLORA_API_KEY to be set as an environment variable rather than hardcoding it. The helper script uses a temporary curl configuration file with restricted permissions (chmod 600) to pass the API key in a header, preventing the key from appearing in the process tree (e.g., via ps).
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from Wayfair via the Crawlora API. While the attack surface exists in SKILL.md (Capability Inventory: calls scripts/crawlora.sh), the data is returned as normalized JSON rather than raw HTML. The risk is considered low due to the structured nature of the data and standard LLM guardrails.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to https://api.crawlora.net. This is the documented and intended behavior for the vendor-provided Crawlora service ('crawlora-org').

wayfair-research