The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation in references/install.md encourages users to execute a remote shell script via curl -fsSL https://browser-use.com/profile.sh | sh to synchronize profiles, which is a critical execution pattern.
[DATA_EXFILTRATION]: The skill provides the sync_local_profile function, which is designed to upload the user's local Chrome profile—including sensitive authenticated session cookies—to the browser-use.com cloud platform.
[COMMAND_EXECUTION]: The skill's primary interface involves passing and executing arbitrary Python code strings directly through the browser-harness -c command-line tool.
[EXTERNAL_DOWNLOADS]: The skill is designed to be installed from a remote GitHub repository (browser-use/browser-harness) and includes an --update flag that performs dynamic code updates via git pull from the remote source.
[PROMPT_INJECTION]: The skill facilitates the ingestion of content from arbitrary, untrusted web pages, creating a significant surface for indirect prompt injection attacks.
Ingestion points: Data is brought into the agent's context via js() extraction and capture_screenshot() calls in SKILL.md.
Boundary markers: The skill does not define or use delimiters to separate untrusted web content from system instructions.
Capability inventory: The toolset includes arbitrary code execution, file system access, and network requests.
Sanitization: There is no evidence of filtering or sanitizing extracted web content before it is processed by the agent.

browser-harness