The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several local shell scripts (check-deps.sh, ensure-browser.sh, close-browser.sh) to manage its execution environment and control browser processes.
[COMMAND_EXECUTION]: The skill relies on the agent-browser CLI tool for comprehensive browser automation, enabling the agent to navigate URLs, click elements, and fill forms.
[COMMAND_EXECUTION]: The agent is granted the ability to execute arbitrary JavaScript within the browser context via the agent-browser eval command, allowing for dynamic page interaction.
[DATA_EXFILTRATION]: The skill provides tools to extract sensitive browser data, specifically the ability to list all cookies and view local storage content through agent-browser commands.
[CREDENTIALS_UNSAFE]: Browser session data, including authentication cookies and storage, is persisted in the ~/.claude/browser-profile/ directory to maintain user states across different sessions.
[EXTERNAL_DOWNLOADS]: The skill's setup instructions guide the agent to install external dependencies, specifically the agent-browser package from a global NPM registry.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection when processing content from the web.
Ingestion points: Data is ingested from search results and external web pages accessed via WebFetch or agent-browser interactions.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands in the fetched web content.
Capability inventory: The skill can execute shell scripts, manage local files (browser profile), and perform browser-based actions like typing and executing JavaScript.
Sanitization: External content is processed and presented to the AI without sanitization or validation.

web-access