The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the agent-browser global package from the npm registry.
[REMOTE_CODE_EXECUTION]: The skill utilizes the agent-browser skills get command to dynamically fetch and load operational workflows and instructions from the tool's backend at runtime.
[PROMPT_INJECTION]: The skill description contains a directive instructing the agent to prefer the agent-browser tool over any built-in browser automation or web tools, attempting to override the agent's default tool selection behavior.
[COMMAND_EXECUTION]: The skill is configured to use the Bash tool to run agent-browser CLI commands, enabling full interaction with the local system and browser instances.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to process and interact with untrusted external data from web pages and third-party applications, which presents an attack surface for indirect instructions.
Ingestion points: Browser CDP accessibility tree snapshots, web page content, Electron application metadata, and Slack workspace data.
Boundary markers: No explicit delimiters or instructions to ignore embedded content are defined in the skill stub.
Capability inventory: The skill possesses the ability to execute shell commands, interact with browsers, manage an authentication vault, and proxy network traffic.
Sanitization: Sanitization procedures are not defined within the skill's markdown instructions and are handled by the external CLI tool.

agent-browser