The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the creation and execution of dynamic code. It instructs the agent to write logic to the ./.bcode/agent-workspace/ directory and subsequently load and run those scripts using dynamic await import() calls with cache-busting parameters.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted external data.
Ingestion points: Untrusted data enters the agent context from external websites (e.g., page titles or DOM content) retrieved through session.Runtime.evaluate calls described in SKILL.md.
Boundary markers: No delimiters or boundary markers are defined to isolate untrusted web content from the agent's instructions.
Capability inventory: The skill provides access to the browser's JavaScript environment, network requests to vendor APIs, and the ability to write to the local file system via project-specific workspaces.
Sanitization: There is no evidence of sanitization, validation, or escaping of the processed web content before it is used by the agent.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to api.browser-use.com and browser-use.com for browser provisioning, authentication, and signup. These domains are official resources belonging to the skill's author and are used for legitimate service operations.

browser-execute