The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill follows secure practices for handling API keys by requiring them in environment variables rather than hardcoding them. The helper script scripts/crawlora.sh further secures the key by writing it to a temporary configuration file with restricted permissions (chmod 600) and passing it to curl via --config, which prevents the key from appearing in process lists (e.g., via ps or top).- [COMMAND_EXECUTION]: The helper script scripts/crawlora.sh implements several defensive measures against command injection and unauthorized access. It uses strict path allowlisting for API routes and validates that the API key format matches an expected character set. It also explicitly rejects the @ character in query parameters to prevent curl from interpreting inputs as local file paths for upload (a common technique for local file disclosure).- [DATA_EXFILTRATION]: Network requests are restricted to the official vendor API endpoint (api.crawlora.net). The script hardcodes the base URL to prevent redirection of sensitive data to attacker-controlled hosts via environment variables.- [INDIRECT_PROMPT_INJECTION]: While the skill ingests external data from Wish, it acts as a passive research tool. The fetched JSON data is intended for information retrieval, and the skill does not expose dangerous capabilities (like file writing or shell execution) that could be exploited by malicious content embedded in product descriptions or reviews.

wish-research