develop-web-game
Warn
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PRIVILEGE_ESCALATION]: The
SKILL.mdfile contains instructions to perform a global package installation usingbun add -g @playwright/mcp@latest. Global installations modify the host environment and typically require elevated permissions. - [DYNAMIC_EXECUTION]: The
scripts/web_game_playwright_client.jsscript dynamically generates a JavaScript shim at runtime (makeVirtualTimeShim) and injects it into the browser context usingpage.addInitScript. It also usespage.evaluateto execute logic constructed as strings within the target page. - [COMMAND_EXECUTION]: The skill workflow relies on executing a local Node.js script (
web_game_playwright_client.js) which launches and manages a Chromium browser process. The script accepts a directory path for screenshots that is used in file system operations without explicit sanitization. - [INDIRECT_PROMPT_INJECTION]: The skill instructions mandate reading a
progress.mdfile to maintain context between agent sessions, creating a surface for instructions embedded in the file to influence agent behavior. - Ingestion points: Content of
progress.mdand game state output fromwindow.render_game_to_text. - Boundary markers: Absent; there are no specific delimiters or warnings to ignore instructions within these data sources.
- Capability inventory: Execution of shell commands via Node.js, file system writes via
fs.writeFileSync, and network interaction through the automated browser. - Sanitization: None; the agent is instructed to read the file content and continue work based on its notes.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install external software including the Bun runtime and the
@playwright/mcppackage from public registries.
Audit Metadata