The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables an agent to process content from external websites, creating a surface for indirect prompt injection where malicious content on a webpage could attempt to influence the agent's behavior.
Ingestion points: The skill uses snapshot -i and get text to retrieve content from web pages and provide it to the agent's context (SKILL.md, references/snapshot-refs.md).
Boundary markers: The instructions do not provide delimiters or guidelines for the agent to treat website-sourced text as untrusted or to ignore embedded instructions.
Capability inventory: The agent has access to agent-browser commands including click, fill, eval (JavaScript execution), and navigation. It also includes capabilities to access local files via file:// URLs and the --allow-file-access flag (SKILL.md, references/commands.md).
Sanitization: No explicit sanitization, filtering, or validation of the content retrieved from websites is defined in the skill instructions.
[COMMAND_EXECUTION]: The skill uses a custom CLI tool agent-browser via Bash. It includes instructions for global installation and management of browser sessions, including state persistence for authentication and proxy configuration (references/commands.md, references/session-management.md, templates/authenticated-session.sh).
[DYNAMIC_EXECUTION]: The skill provides an eval command (references/commands.md) which allows the execution of JavaScript directly within the browser context. While intended for automation tasks, this represents a dynamic execution capability.

agent-browser