Skip to content

Latest commit

 

History

History
45 lines (37 loc) · 2.43 KB

File metadata and controls

45 lines (37 loc) · 2.43 KB

AGENTS Instructions — Omni-Workspace Root

CRITICAL: READ docs/UNIVERSAL_LLM_INSTRUCTIONS.md FIRST.

1. Agent Overview

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.

2. Multi-Agent Collaboration & Handoff

  • 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.

3. Model-Specific Overrides

Do not duplicate instructions. Refer to the model-specific files for unique capabilities and rules:

4. Version Management Protocol

  • Single source of truth: VERSION.md contains 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)

5. Documentation Protocol

  • 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

6. Deployment Protocol

  • See DEPLOY.md for complete deployment instructions
  • Always use BACKEND_FORCE_TAR=1 for backend deploys
  • Use npx vite build (not npm run build) to avoid Windows git errors
  • Verify deployment with health checks before proceeding

7. Safety Rules

  • DO NOT taskkill any 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