godot-ai-staff-stresstest

# Godot TDD Multi-Agent Template A port of the [ai-staff-template](../../README.md) TDD Agentic Framework to Godot 4.6.1.stable game development, wired to the [GDAI MCP server](https://gdaimcp.com/) for fully autonomous Red-Green-Refactor cycles. > **Status:** Experimental — scaffold complete, awaiting MCP configuration and Godot project initialization. > See `BOOTSTRAPPING.md` to get started. --- ## What This Is This template gives you a structured AI staff for Godot development. Each agent has a defined role, a mandate, and hard constraints. They collaborate through a TDD pipeline where **no feature is implemented until a failing test proves it doesn't exist yet**, and **no feature is shipped until MCP confirms the test passes**. The GDAI MCP server is the backbone. It gives agents programmatic control over the Godot editor — creating scenes, editing GDScript, running the game, reading debugger output, capturing screenshots, and simulating input — closing the TDD loop without a human in the middle. --- ## The Staff | Name | Role | Mandate | |---|---|---| | **VERA** | Orchestrator | Task decomposition, fun-factor ROI, MCP oversight | | **CASS** | Test Architect | Writes failing GUT tests before any scene is touched | | **AXEL** | Gameplay Architect | Implements GDScript logic — only what the test demands | | **WREN** | Scene Designer | Builds `.tscn` hierarchies and UI scenes | | **KAI** | Integration Agent (The Bridge) | Owns signal contracts, runs MCP E2E verification | | **RIVEN** | QA Agent (The Adversary) | Adversarial input simulation, frame time, threat assessment | | **ECHO** | Archivist | GDD alignment, signal registry, decision log | | **SOREN** | Systems Analyst | Cycle time, fun-factor ROI, operational debt | Full persona definitions are in `.agents/personas/`. --- ## The TDD Pipeline Every feature follows this sequential pipeline. No step begins until the previous one signals `READY`. ``` 1. RED Test Architect writes a failing GUT test. MCP runs the scene → debugger confirms FAILED. 2. GREEN Gameplay Architect + Scene Designer implement only what the test demands. MCP runs the scene → debugger confirms PASSED. 3. BRIDGE Integration Agent verifies signal contracts. MCP runs a full game flow + captures screenshot. Issues Verification: PASS or FAIL with evidence. 4. QA QA Agent simulates adversarial input via MCP. Checks frame time against budget. Files a Threat Assessment in memory/threat_log.md. 5. ARCHIVE Systems Analyst drafts post-mortem. Archivist updates context_snapshot.md and signal_registry.md. Session is closed. ``` The equivalent in the parent web framework: | Web Stack | Godot Equivalent | |---|---| | `pytest` | GUT (`extends GutTest`) | | `Playwright` | GDAI MCP (run game + screenshot) | | OpenAPI ↔ Zod parity | Signal registry contracts | | Backend Architect | Gameplay Architect | | Frontend Specialist | Scene Designer | | Security Agent | QA Agent | --- ## Directory Structure ``` godot-ai-staff-stresstest/ ├── README.md ← You are here ├── BOOTSTRAPPING.md ← Start here to set up a real project └── .agents/ ├── personas/ │ ├── orchestrator.md │ ├── test_architect.md │ ├── gameplay_architect.md │ ├── scene_designer.md │ ├── integration_agent.md │ ├── qa_agent.md │ ├── archivist.md │ └── systems_analyst.md ├── mandates/ │ ├── global_standards.md ← GDScript rules, MCP protocol │ ├── gameplay_mandate.md ← Signals, state machines, physics │ ├── scene_mandate.md ← Ownership, PackedScene contracts │ ├── test_mandate.md ← GUT conventions, TDD Triple-Jump │ └── delegation_protocol.md ← The full pipeline └── memory/ ├── context_snapshot.md ← Current stable project state ├── active_sprint.md ← Sprint tasks and frame budget ├── decision_log.md ← ADR-format architecture decisions ├── signal_registry.md ← Inter-scene signal contracts ├── threat_log.md ← QA Agent assessments ├── operational_debt.md ← Debt impact matrix └── session_post_mortem.md ← Efficiency audits per session ``` --- ## Key Design Decisions **Why GUT?** GUT (Godot Unit Test) is the most widely adopted Godot test framework. It's GDScript-native, supports signal assertions (`assert_signal_emitted`), and can be run headlessly — compatible with MCP-driven execution. See `memory/decision_log.md` for the full rationale. **Why GDAI MCP over manual verification?** Without MCP, agents can write GDScript but can't confirm it runs. MCP closes the Red-Green loop autonomously: the Test Architect writes a failing test, MCP confirms `FAILED`, the Gameplay Architect implements, MCP confirms `PASSED`. No human execution step in the middle. **Why signal contracts over direct node references?** Godot scenes that reach into each other's node trees via `get_node("/root/...")` create tight coupling that breaks when scenes are refactored. Signal contracts enforced by the Integration Agent keep scenes independently runnable and refactor-safe. --- ## Prerequisites - Godot 4.6.1.stable - GDAI MCP plugin (one-time $19 — [gdaimcp.com](https://gdaimcp.com/)) - GUT plugin ([github.com/bitwes/Gut](https://github.com/bitwes/Gut)) - Claude Code with MCP configured See `BOOTSTRAPPING.md` for the full setup walkthrough.