The Agent Skills Directory

[PROMPT_INJECTION]: The skill is explicitly designed to bypass organizational compliance and safety training protocols by directly faking 'passed' status within the SCORM 1.2 LMS API rather than completing the material.
[COMMAND_EXECUTION]: The skill uses shell commands (head, tail) to read the DevToolsActivePort file from local browser user data directories across multiple operating systems. This file contains sensitive WebSocket identifiers that allow full remote control of the browser session.
[COMMAND_EXECUTION]: The skill utilizes the agent-browser tool to execute arbitrary JavaScript strings within the context of the user's authenticated Workday browser session to manipulate internal application state.
[COMMAND_EXECUTION]: The skill employs dynamic JavaScript execution patterns, such as using eval() to interact with the window.API and DS.playerGlobals objects in the browser to bypass UI constraints.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests untrusted data from external web pages (Workday course content) through element snapshots and screenshots.
Ingestion points: Browser tab content processed via agent-browser snapshot -i and screenshot (SKILL.md).
Boundary markers: Absent; there are no instructions to ignore embedded content within the ingested DOM.
Capability inventory: Shell command execution and arbitrary JavaScript execution in the browser (SKILL.md).
Sanitization: Absent; ingested element trees and text are not sanitized before being used to inform agent actions.

workday-training