The Agent Skills Directory

[SAFE]: The skill is designed to perform systematic UI walkthroughs and audit applications for UX improvements. It instructions the agent to use built-in browser automation tools to capture screenshots across multiple viewports (desktop, mobile, tablet).
[COMMAND_EXECUTION]: The skill employs vendor-specific CLI and MCP tools (helpmetest, helpmetest_run_test, run_interactive_command) to manage authentication states and perform live app navigation. This is standard functionality for the helpmetest ecosystem.
[DATA_EXFILTRATION]: While the skill uploads screenshot data via helpmetest_upload, this action is restricted to the vendor's own infrastructure for the purpose of generating review artifacts and is not considered unauthorized exfiltration.
[PROMPT_INJECTION]: The skill references a character configuration file (.helpmetest/SOUL.md) to define the agent's persona. This is a local configuration mechanism and does not attempt to bypass platform safety guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it processes data from local files and web pages.
Ingestion points: External content is ingested from the .helpmetest/SOUL.md configuration file and from rendered text/UI elements on web pages visited during the walkthrough.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded prompts within the audited pages or configuration files.
Capability inventory: The skill can execute browser commands via run_interactive_command, run tests via helpmetest_run_test, and save data via helpmetest_upsert_artifact.
Sanitization: No explicit sanitization or validation steps are described for the data retrieved from the target application's UI or the character file.

ui-review