The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Linux installation instructions utilize a pattern of piping remote content directly into a system command with elevated privileges: `wget -q -O
https://dl-ssl.google.com/linux/linux_signing_key.pub | sudo apt-key add -.- **[DATA_EXFILTRATION]:** The skill provides direct access to browser cookies through endpoints like GET /api/v1/cookies/browser. This allows for the extraction of sensitive session tokens and authentication data.- **[COMMAND_EXECUTION]:** The documentation includes multiple commands requiring sudofor package management and repository configuration. It also instructs the agent to execute system-level troubleshooting tools includingpkill, lsof, netstat, and tail for log analysis.- **[PROMPT_INJECTION]:** The AI exploration feature (POST /api/v1/ai-explore/start`) ingests data from external websites which could contain malicious instructions.
Ingestion points: Task descriptions and live content from arbitrary URLs during automated browsing.
Boundary markers: None identified.
Capability inventory: Browser automation tools (navigate, click, type), script generation, and LLM configuration management.
Sanitization: No methods for sanitizing or escaping website content before processing are described.- [PROMPT_INJECTION]: The skill exposes an API to modify the platform's system prompts (PUT /api/v1/prompts/:id), providing a mechanism to override core AI behavior or safety constraints.- [CREDENTIALS_UNSAFE]: The skill facilitates the management and storage of LLM API keys (e.g., OpenAI, Anthropic) via the /api/v1/llm-configs endpoint, representing a risk for credential exposure.- [EXTERNAL_DOWNLOADS]: Fetches software packages and security keys from Google's official distribution servers for Chrome installation.

browserwing-admin