The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a helper script scripts/crawlora.sh to execute curl commands. The script includes robust validation for input paths and environment variables, and it correctly uses --data-urlencode to prevent command injection through query parameters. No arbitrary shell execution or unsafe interpolation was detected.
[DATA_EXFILTRATION]: Network requests are restricted to the vendor's API at api.crawlora.net. The skill avoids exposing the API key in the process list by writing it to a temporary file with restricted permissions (chmod 600) and using curl's --config flag, which is a security best practice for handling secrets in shell scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests product data and reviews from an external API, which represents a standard surface for indirect prompt injection where untrusted external content enters the agent's context.
Ingestion points: API responses for product descriptions and customer reviews processed in SKILL.md via scripts/crawlora.sh.
Boundary markers: Not explicitly implemented in the shell script output, as it returns raw JSON.
Capability inventory: Network access via curl and shell execution restricted to the scripts/crawlora.sh helper.
Sanitization: Input validation is present for the API key and request paths to ensure they conform to expected formats.

nike-research