The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to execute shell commands that launch third-party applications such as Slack, Visual Studio Code, Discord, Figma, Notion, and Spotify with the --remote-debugging-port flag. This enables external control via the Chrome DevTools Protocol.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external applications using snapshot and text extraction tools, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Data is brought into the agent context via agent-browser snapshot -i, agent-browser get text, and agent-browser snapshot --json commands in SKILL.md.
Boundary markers: The instructions do not define delimiters or specific markers to differentiate untrusted application data from agent instructions.
Capability inventory: The skill includes capabilities to interact with the environment through agent-browser click, agent-browser fill, and agent-browser keyboard commands.
Sanitization: There is no evidence of sanitization, filtering, or validation of the content retrieved from the target applications before it is processed by the agent.

electron