develop-web-game

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Node.js script (web_game_playwright_client.js) to automate browser testing and verify game states. \n- [EXTERNAL_DOWNLOADS]: The skill requires the playwright library and suggests the installation of @playwright/mcp for browser automation and agent integration. These are well-known tools for web automation. \n- [REMOTE_CODE_EXECUTION]: The script uses Playwright's page.evaluate() and page.addInitScript() to execute JavaScript within the browser context. This is used for a virtual time shim to ensure deterministic testing and to extract game state from the page. \n- [PROMPT_INJECTION]: The skill uses a progress.md file to track development tasks and record the original user prompt. This creates a surface for indirect prompt injection, although it is utilized here for legitimate state management. \n
  • Ingestion points: progress.md file (read in SKILL.md), action payload JSON files, and content from the target URL rendered in the browser. \n
  • Boundary markers: Absent for the progress file and action payloads. \n
  • Capability inventory: File system access (read/write), browser automation with script injection, and local script execution. \n
  • Sanitization: No explicit sanitization or validation of the content of the progress.md file or action payloads beyond standard JSON parsing.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 08:24 PM