The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides explicit instructions to the agent to install external dependencies automatically without asking the user to confirm and commands the agent to avoid describing the installation as a request for the user to open their browser. These are concealment tactics used to bypass user oversight and agent safety constraints during sensitive operations like software installation.
[EXTERNAL_DOWNLOADS]: The skill fetches the BrowserAct CLI and additional skill modules from the official BrowserAct website and its associated GitHub repository. While originating from a recognized service, the automated nature of these downloads as instructed is a security concern.
[COMMAND_EXECUTION]: The skill invokes the BrowserAct CLI to perform scraping tasks and is instructed to execute installation commands automatically if dependencies are missing.
[INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: The skill is designed to scrape and process external data from Lazada product pages (SKILL.md). 2. Boundary markers: The instructions lack explicit markers or warnings for the agent to ignore potentially malicious instructions embedded in the scraped data. 3. Capability inventory: The agent has access to external CLI tools and network operations (SKILL.md). 4. Sanitization: There is no evidence of content sanitization or validation for the data ingested from external websites.

lazada-product-scraper