The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it ingests content from external websites.
Ingestion points: External data is ingested through agent-browser open, get text body, and get html main commands in SKILL.md.
Boundary markers: Absent; there are no delimiters or instructions to ignore commands found within the web content.
Capability inventory: The skill possesses extensive capabilities including agent-browser eval, click, fill, and upload (as seen in SKILL.md).
Sanitization: Absent; the content is not filtered or sanitized before being processed by the agent.
[DATA_EXFILTRATION]: The skill can access sensitive browser session data.
Evidence: The commands agent-browser cookies get and agent-browser storage local get enable retrieval of potentially sensitive authentication and session information.
[COMMAND_EXECUTION]: The workflow relies on executing the agent-browser CLI tool to perform all browser interactions.
[EXTERNAL_DOWNLOADS]: The skill suggests the command agent-browser install to download browser binaries from external sources if they are missing.
[REMOTE_CODE_EXECUTION]: Includes an eval command that allows the execution of arbitrary JavaScript within the browser's execution context.

browse