The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from external app stores, such as app descriptions and user reviews.
Ingestion points: Data is fetched from Apple App Store, Google Play, and Chrome Web Store via API calls made in scripts/crawlora.sh.
Boundary markers: The skill returns data in JSON format but does not add specific delimiters or instructions to the LLM to ignore potentially malicious text within the retrieved descriptions or reviews.
Capability inventory: The skill uses curl to interface with the Crawlora API; no file-write or further network capabilities are present in the provided scripts.
Sanitization: While query parameters are URL-encoded, the descriptive text and review body content from the store listings are returned as provided by the upstream source without modification.
[COMMAND_EXECUTION]: The skill relies on a bash helper script scripts/crawlora.sh to facilitate API requests.
The script implements defensive programming practices including strict route validation via an allow-list of paths and anchored regular expressions to prevent path traversal or unauthorized API access.
It validates the format of the CRAWLORA_API_KEY to ensure it does not contain characters that could disrupt the configuration file generation or shell environment.
It uses a temporary curl configuration file (with appropriate permissions and cleanup) to pass the API key, ensuring the secret never appears in the system's process list.
[EXTERNAL_DOWNLOADS]: Communicates with the vendor's API at api.crawlora.net to retrieve research data.
The base URL is hardcoded in the script to prevent redirection via environment variables.
The communication is restricted to GET requests for the documented research routes only.

app-store-research