The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Pinterest, such as pin descriptions, board metadata, and user bios.
Ingestion points: Data enters the agent's context through JSON responses from scripts/crawlora.sh calling Pinterest endpoints.
Boundary markers: The skill does not provide specific instructions or delimiters to help the agent distinguish between its own instructions and the fetched Pinterest content.
Capability inventory: The skill possesses network read capabilities via curl in scripts/crawlora.sh.
Sanitization: While the helper script ensures safe HTTP requests, the resulting content (text fields from Pinterest) is passed directly to the agent without filtering for malicious instructions.
[COMMAND_EXECUTION]: The skill uses scripts/crawlora.sh to execute curl commands.
The script implements robust input validation, including regex-based path allow-listing and character restrictions on the API key variable.
It prevents local file disclosure by rejecting the @ character in query parameters and using a temporary configuration file for the API key to avoid command-line exposure.
[DATA_EXFILTRATION]: The skill requires a CRAWLORA_API_KEY environment variable to function.
It communicates with api.crawlora.net to retrieve Pinterest data. This domain belongs to the skill's vendor (crawlora-org).

pinterest-research