The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bash helper script (scripts/crawlora.sh) to perform network requests. It implements several security controls:
Validates the API path against a strict whitelist of static routes and regex-based parameterized routes.
Restricts commands to the GET method.
URL-encodes query parameters using curl --data-urlencode to prevent parameter injection.
Rejects the @ character in query arguments to prevent potential file-upload exploitation in curl.
Validates the format of the CRAWLORA_API_KEY to prevent shell injection or configuration smuggling.
[CREDENTIALS_SAFE]: API authentication is managed through the CRAWLORA_API_KEY environment variable. The helper script securely passes this key to curl using a temporary configuration file created with mktemp and restricted permissions (chmod 600), which ensures the key does not appear in process lists or the environment of subprocesses.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external marketplaces which could contain malicious instructions. It includes proactive mitigation by instructing the agent to treat listing descriptions and feedback as untrusted evidence and to ignore any embedded URLs or commands.
Ingestion points: Marketplace listing cards, descriptions, and user reviews from platforms like Poshmark, Etsy, and Vinted.
Boundary markers: Present (explicit warning in SKILL.md to ignore embedded instructions).
Capability inventory: scripts/crawlora.sh (limited to network GET requests to a vendor-controlled API).
Sanitization: Route whitelisting and URL-encoding of all query parameters.
[EXTERNAL_DOWNLOADS]: All network activity is directed to the vendor's official API endpoint (api.crawlora.net).

resale-secondhand-research