The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses browser tools such as page.extract_text and page.get_html to ingest content from external websites. This creates a surface for indirect prompt injection, where a malicious website could provide instructions that manipulate the agent's behavior. * Ingestion points: page.extract_text and page.get_html * Capability inventory: Browser interaction tools (click, type) and shell command execution * Boundary markers: None identified in the provided constraints * Sanitization: No explicit content filtering or validation is mentioned.
[DATA_EXFILTRATION]: The skill configures browser automation to use a specific user profile path (~/.chrome-cdp-profile). While intended for session persistence, this path is sensitive as it stores user authentication cookies, history, and other private session data.
[COMMAND_EXECUTION]: The skill utilizes shell commands (sleep $((RANDOM % 26 + 5))) to implement randomized delays between automation steps to evade detection mechanisms.

skill-constraints