competitor-intelligence
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a local Bash script,
scripts/crawlora.sh, to interact with the Crawlora API. The script includes significant defensive measures, such as validating the API key format, using a temporary configuration file to keep credentials out of the process list, allowlisting specific API paths and methods, and preventing local file disclosure via curl's file-access syntax. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to fetch and process data from external websites (e.g., product pages, reviews), which represents an untrusted data ingestion surface.
- Ingestion points: Untrusted content enters the agent's context through the JSON output of
scripts/crawlora.shfor various scrape and search endpoints described inreference/endpoints.md. - Boundary markers: The skill instructions guide the agent to categorize the data (claims vs. observations) but do not define strict delimiters or include specific instructions to ignore embedded prompts within the retrieved data.
- Capability inventory: The skill provides the agent with network access capabilities through the provided API helper script.
- Sanitization: While the script validates tool-level parameters to prevent system-level attacks, there is no mechanism to sanitize the natural language content of the scraped pages to prevent potential adversarial instructions from influencing the AI model's logic.
Audit Metadata