The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute a local CLI tool named jb (located at bin/jb). This tool is used to launch and drive a Chromium browser session, manage goals, and handle interactive text requests during browsing.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to navigate and extract data from external, untrusted websites. This creates a surface for indirect prompt injection where malicious content on a target page could attempt to influence the agent's logic.
Ingestion points: Data is ingested from web page content, including titles, body text, and links, during the 'hunt' for information.
Boundary markers: The Safety section explicitly defines page content as untrusted and prohibits actions like logins or purchases without explicit user approval.
Capability inventory: The skill can execute shell commands via the jb launcher and has the ability to read local files and environment configuration.
Sanitization: The skill relies on the agent's interpretation of the Safety instructions and the underlying tool's behavior rather than programmatic sanitization of the webpage content.
[DYNAMIC_EXECUTION]: The bin/jb script is a Node.js launcher that dynamically calculates the path to the actual jb binary (../../../bin/jb) using realpathSync and join. This resolution happens at runtime based on the script's location on the filesystem.
[CREDENTIALS_UNSAFE]: The skill documentation references a configuration file at ~/.jb/config.json and explicitly notes that it contains keys. While the instructions warn the agent not to print these keys, the skill provides the mechanism to interact with this sensitive local configuration.

jb-browser