The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the installation of a global Node.js package from the official NPM registry.
Evidence: npm i -g agent-browser && agent-browser install (SKILL.md).
[COMMAND_EXECUTION]: The skill uses shell commands via the Bash tool to perform all browser automation, desktop application interaction, and configuration tasks.
Evidence: The allowed-tools frontmatter grants access to the agent-browser CLI via Bash (SKILL.md).
[DYNAMIC_EXECUTION]: The skill employs a dynamic instruction-loading pattern where operational workflows and command references are fetched at runtime using the CLI.
Evidence: Commands like agent-browser skills get core and agent-browser skills get electron are used to load specialized instructions after installation (SKILL.md).
[INDIRECT_PROMPT_INJECTION]: As a browser automation tool, the skill has a significant surface for indirect prompt injection when interacting with external web content or communication platforms like Slack.
Ingestion points: Web pages, Electron application content, and Slack workspace data accessed via Chrome/CDP (SKILL.md).
Boundary markers: None identified in the discovery stub; the agent is directed to load external workflows for specific guidance.
Capability inventory: The tool can navigate pages, fill forms, click buttons, extract data, and interact with authenticated sessions via an "authentication vault" (SKILL.md).
Sanitization: No specific sanitization or filtering logic is described in this high-level discovery stub.

agent-browser