CRITICAL: READ docs/UNIVERSAL_LLM_INSTRUCTIONS.md FIRST.
This workspace is a multi-agent orchestration environment. All agents (Gemini, Claude, GPT, Copilot, Jules) must follow the core protocols defined in the universal instructions.
- Gemini CLI: Performance, large context operations, workspace-wide synchronization, and deep codebase analysis.
- Claude: Architecture, planning, system-wide design, and maintaining extreme-detail global documentation.
- GPT: Specific implementation details, unit tests, fast coding, and autonomous algorithmic execution.
- Google Jules: Continuous development, branch maintenance, and periodic sweeping updates.
Do not duplicate instructions. Refer to the model-specific files for unique capabilities and rules:
- Single source of truth:
VERSION.mdcontains only the version string (e.g.,2.1.73) - Bump on every deploy: Update
VERSION.md,package.json,MainMenuScene.ts,server/index.js - Commit message: Must reference version number (e.g.,
v2.1.73: description) - Changelog: Every version bump gets an entry in
CHANGELOG.md - Display: Version is shown in the main menu UI (bottom-right corner)
- Session start: Read AGENTS.md, VISION.md, MEMORY.md, ROADMAP.md, TODO.md, HANDOFF.md
- During: Update docs when making significant changes
- Session end: Update HANDOFF.md, MEMORY.md, ROADMAP.md, TODO.md, CHANGELOG.md
- Quality: Documentation must be in "FULL extreme detail" — comprehensive, thorough, actionable
- See DEPLOY.md for complete deployment instructions
- Always use
BACKEND_FORCE_TAR=1for backend deploys - Use
npx vite build(notnpm run build) to avoid Windows git errors - Verify deployment with health checks before proceeding
- DO NOT
taskkillany processes — this kills active sessions - DO NOT force push or overwrite working code
- DO merge intelligently, preserving all features
- DO commit and push regularly between features
- DO keep going autonomously without pausing for confirmation