The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a helper script (scripts/crawlora.sh) that executes curl to interact with the vendor's API. This script implements several security best practices:
Hardcoded Base URL: The API base is hardcoded to https://api.crawlora.net/api/v1, preventing redirection to attacker-controlled hosts.
Path Allow-listing: The script enforces a strict allow-list for API routes using case matching and anchored regular expressions, preventing unauthorized API access or path traversal.
Secure Credential Handling: The CRAWLORA_API_KEY is validated for format and passed to curl via a temporary configuration file with restricted permissions (chmod 600), ensuring the key is not exposed in the process command line.
Input Sanitization: Query parameters are URL-encoded using curl's --data-urlencode flag, and the @ character is blocked to prevent unauthorized file reads/uploads via curl's built-in syntax.
[INDIRECT_PROMPT_INJECTION]: The skill fetches data from an external source (Rover marketplace data via the Crawlora API), creating a potential ingestion surface for indirect prompt injection.
Ingestion points: API response data retrieved via scripts/crawlora.sh from the vendor's infrastructure.
Boundary markers: The skill instructions suggest piping output to jq, which encourages structured data processing rather than direct execution of returned content.
Capability inventory: The skill is restricted to making GET requests to specific vendor endpoints; it does not have the capability to write files, modify system state, or execute arbitrary shell commands based on the external data.
Sanitization: The helper script ensures that all user-supplied query parameters are properly URL-encoded before the request is made.

pet-services-research