The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages an authenticated Google browser profile stored at ~/.uxc/webmcp-profile/google. This directory contains sensitive session data and cookies required for the agent to act on behalf of the user on Google services.
[COMMAND_EXECUTION]: The skill includes shell scripts (scripts/ensure-links.sh) used to configure the execution environment. This script executes a secondary script located in a sibling skill directory (skills/webmcp-bridge/scripts/ensure-links.sh) to setup the CLI tool.
[EXTERNAL_DOWNLOADS]: The skill instructions require the installation of browser binaries via npx playwright install. These downloads are performed from well-known sources associated with Microsoft's Playwright project.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data retrieved from external sources (Google Search results and Gemini responses).
Ingestion points: Data is ingested through search.web results and gemini.chat responses (SKILL.md).
Boundary markers: None explicitly defined in the instructions to separate search results from agent instructions.
Capability inventory: The skill can execute shell commands and automate browser interactions.
Sanitization: The skill relies on structured JSON output for tool responses, which provides a layer of data isolation.

google-webmcp