The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the agent-browser command-line tool to perform automated browser actions such as navigation, interaction, and data extraction.
[EXTERNAL_DOWNLOADS]: The documentation references npx @claude-flow/cli, which dynamically downloads and executes packages from the npm registry.
[CREDENTIALS_UNSAFE]: Includes commands to save and load browser session states (e.g., agent-browser --session nav state save auth.json). These files typically contain sensitive cookies and authentication tokens that could be exposed if the local environment is compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection attacks where a malicious website could embed instructions designed to manipulate the agent's behavior.
Ingestion points: Data enters the agent context through agent-browser snapshot, get text, get html, and get title commands which parse external web content.
Boundary markers: The skill does not define specific delimiters or instructions to prevent the agent from obeying commands found within the accessibility tree or page text.
Capability inventory: The agent has extensive control over the browser, including the ability to navigate to new URLs, fill forms, click elements, and persist session states.
Sanitization: No sanitization or filtering of the ingested web content is described, though the use of an accessibility tree (snapshot) may reduce some surface area compared to raw HTML.

browser