The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions allow the agent to identify and execute shell commands to start development servers (e.g., npm run dev) by parsing project files such as package.json. This grants the agent the ability to trigger arbitrary processes defined within the codebase.\n- [CREDENTIALS_UNSAFE]: The agent is instructed to read .env and other environment configuration files to detect port and host settings. These files typically contain sensitive API keys and secrets, creating a risk of data exposure if the agent incorporates this information into its context or reports.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection.\n
Ingestion points: The agent interacts with live web applications, reading DOM content, console logs, and network traffic (SKILL.md).\n
Boundary markers: No explicit delimiters or instructions are provided to the agent to isolate instructions embedded within the processed web content.\n
Capability inventory: The skill allows the agent to execute shell commands for starting servers and to modify source code in 'fix mode' (SKILL.md).\n
Sanitization: There is no evidence of sanitization or filtering applied to the external web data before it is processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill references downloading and installing browser dependencies via npx when they are missing from the host environment.

fieldtest