The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the output of local Python scripts directly in a shell using eval "$(python scripts/xxx.py)". This pattern is inherently dangerous as it treats script output as executable code without validation. Additionally, it references the use of stealth-extract, an external CLI tool, for processing website content.
[PROMPT_INJECTION]: The skill has a high surface area for Indirect Prompt Injection (Category 8). It navigates to and extracts content from arbitrary third-party product websites to identify email addresses and descriptions. Malicious instructions embedded on these websites could hijack the agent's behavior.
Ingestion points: scripts/extract-website-content.py (reads bodyText and innerHTML from external sites).
Boundary markers: Absent. No delimiters or warnings are used to differentiate scraped content from the agent's core instructions.
Capability inventory: Includes shell execution (eval), browser control (browser-act), and navigation to external domains.
Sanitization: None detected beyond simple whitespace removal and truncation to 5000 characters.
[DATA_EXFILTRATION]: The core functionality includes automated email harvesting and lead generation from external websites. While transparent about this purpose, it automates the collection of contact data which is then aggregated and returned in structured format to the agent context.
[EXTERNAL_DOWNLOADS]: Automated scanners flagged specific domains in the skill as malicious and identified potential 'HttpRequest' infection patterns in the instructions. While the domains appear to be placeholders, the skill's primary function is to navigate and download data from a vast number of untrusted external sites during the scraping process.

producthunt-launches