The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation instructs the user to pipe a remote shell script from an external, non-trusted domain directly into the system shell for profile synchronization.
Evidence in references/quickstart.md: curl -fsSL https://browser-use.com/profile.sh | sh
Evidence in references/sessions.md: curl -fsSL https://browser-use.com/profile.sh | sh
[EXTERNAL_DOWNLOADS]: The skill recommends the installation of various software development kits and command-line interfaces from public package registries (PyPI and NPM).
Evidence in references/quickstart.md: pip install browser-use-sdk and npm install browser-use-sdk
Evidence in references/guides/subagent.md: uv pip install 'browser-use[cli]'
[COMMAND_EXECUTION]: The skill documents the use of a local CLI tool that performs browser automation actions, including the ability to execute arbitrary JavaScript within a browser context.
Evidence in references/guides/tools-integration.md: Commands such as browser-use open, browser-use click, and browser-use eval which allow for interaction with web environments.
[PROMPT_INJECTION]: The skill facilitates an automated browsing environment that ingests untrusted data from websites which can be used to influence agent behavior through indirect prompt injection.
Ingestion points: browser-use state and browser-use get text allow the agent to read external web content from any URL.
Boundary markers: The documentation does not specify the use of strong delimiters or clear instructions to the agent to ignore embedded commands within the processed web data.
Capability inventory: The agent has the capability to perform secondary network requests, navigate to new domains, and simulate user inputs based on the content it reads.
Sanitization: There is no evidence of content sanitization or filtering to prevent malicious web-based instructions from being executed by the agent.

cloud