The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides explicit instructions for the agent to extract and display sensitive authentication data. This includes capturing Authorization and apikey headers from network requests (why-is-this-request-failing.md), reading all browser cookies, and retrieving session tokens from localStorage (whats-actually-in-storage.md). The decode-a-jwt-without-leaving-the-loop.md playbook provides a JavaScript snippet specifically to extract and decode these tokens.
[INDIRECT_PROMPT_INJECTION]: The skill exposes the agent to indirect prompt injection by design, as it visits external URLs and ingests untrusted data from those pages.
Ingestion points: Data is ingested via browser_console (logs), browser_network (request/response data), and browser_goto (page content).
Capability inventory: The skill utilizes high-impact tools including browser_eval for code execution, browser_fill for input, and browser_click for interaction.
Boundary markers: No boundary markers or instructions are present to warn the agent about potentially malicious instructions embedded in the web data.
Sanitization: The skill lacks mechanisms to sanitize or filter the data retrieved from external web pages before it is processed.
[DYNAMIC_EXECUTION]: The skill frequently uses browser_eval to run JavaScript code generated at runtime for inspecting browser state, validating form fields, and decoding authentication tokens.

web-debug