# What is the Collaborative Protocol in Claude Code Game Studios? Question → Options → Decision → Draft → Approval

> Understand the Claude Code Game Studios collaborative protocol: Question Options Decision Draft Approval. Ensure transparent, user-controlled, traceable agent interactions before file modifications.

- Repository: [Donchitos/Claude-Code-Game-Studios](https://github.com/Donchitos/Claude-Code-Game-Studios)
- Tags: architecture
- Published: 2026-04-16

---

**The collaborative protocol is a mandatory five-step workflow—Question, Options, Decision, Draft, and Approval—that ensures every agent interaction in Claude Code Game Studios remains transparent, user-controlled, and traceable before any file modification occurs.**

The Donchitos/Claude-Code-Game-Studios repository implements a disciplined user-driven architecture where AI agents cannot unilaterally modify code or design documents. Instead, every interaction follows the **collaborative protocol**, a structured loop defined in [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md) (lines 46-48) that gathers context, presents alternatives, and requires explicit user approval before writing to the repository.

## The Five Stages of the Collaborative Protocol

### 1. Question: Clarifying Intent

Every agent interaction begins with focused clarifying questions to gather missing context. This stage guarantees that the user’s vision, constraints, and goals are fully understood before any suggestion is made.

In `CCGS Skill Testing Framework/agents/specialists/game-designer.md`, the protocol is implemented via the `AskUserQuestion` directive:

```markdown
AskUserQuestion:
  questions:
    - question: "What core fantasy should the combat system convey?"
      header: "Combat Vision"
      options:
        - label: "High-Octane Action"
          description: "Fast reflexes, flashy combos"
        - label: "Tactical Chess"
          description: "Turn-based, deep positioning"

```

This YAML structure forces the agent to suspend automation until the user provides directional clarity.

### 2. Options: Presenting Concrete Trade-offs

Once context is established, the agent must present **2-4 concrete options**, each with explicit trade-offs, examples, and alignment to the project’s pillars. This prevents the agent from guessing a single solution and gives the user a clear choice set.

As documented in [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md) (lines 49-90), options are formatted with clear pros and cons:

```markdown
**Option A: Full Randomized Discovery**
- ✅ Maximum exploration reward
- ❌ Can feel arbitrary without skill expression

**Option B: Ingredient Hint System**
- ✅ Rewards observation and deduction
- ❌ Less pure discovery magic

**Option C: Hybrid (Discovery + Hints)**
- ✅ Accessible to multiple player types
- ❌ More complex to implement

```

### 3. Decision: User Authority

The user selects the preferred option or refines it. The system never makes unilateral decisions. According to the protocol definition in [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md), this stage is where the user retains final authority over the design direction.

Example interaction:

```text
User: "Option C feels right. Let's set X = 5 failures to unlock hints."

```

### 4. Draft: Creating Visible Artifacts

Based on the chosen option, the agent creates a draft artifact—whether a design paragraph, formula, code snippet, or file skeleton. This early visibility allows rapid feedback and correction before any permanent change.

From the collaborative design documentation, a draft might look like:

```markdown

### Core Loop

1. Player selects 2-4 ingredients from inventory
2. System checks against recipe database
3. If match → Success, learn recipe, grant full XP
4. If no match → Failure, consume materials, grant 15% XP
5. After 5 failures → unlock hint system for those items

```

### 5. Approval: The Write Gate

The final stage enforces a safety gate. The agent asks, **“May I write this to \<filepath\>?”** Only after an explicit **“Yes”** does the system modify the repository. This pattern is referenced in [`docs/WORKFLOW-GUIDE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/WORKFLOW-GUIDE.md) (lines 72-77) as the mandatory conclusion to every workflow.

Example exchange:

```text
Agent: "May I write this to design/gdd/crafting-system.md?"
User: "Yes"

```

Hooks in [`.claude/statusline.sh`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/.claude/statusline.sh) programmatically enforce this gate, ensuring repository safety.

## Implementation Across the Repository

The collaborative protocol is not merely documentation; it is embedded in the agent architecture:

- **[`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md)** (lines 46-90): Contains the canonical definition and an illustrated end-to-end example of the full loop.

- **[`docs/WORKFLOW-GUIDE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/WORKFLOW-GUIDE.md)** (lines 72-77): References the protocol within the broader project workflow, ensuring all contributors follow the same discipline.

- **[`CLAUDE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/CLAUDE.md)** (root): Summarizes the agent architecture and reiterates the collaborative pattern for system prompts.

- **`CCGS Skill Testing Framework/agents/specialists/*.md`**: Concrete skill files that implement `AskUserQuestion` patterns and write-gate prompts.

- **[`.claude/statusline.sh`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/.claude/statusline.sh)**: Enforces the approval gate before file system modifications occur.

## Summary

- **The collaborative protocol** mandates five sequential stages: Question, Options, Decision, Draft, and Approval.
- **User control** is preserved throughout; agents cannot proceed without explicit user selection and final written confirmation.
- **Source of truth** resides in [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md), with practical implementations in specialist agent files under `CCGS Skill Testing Framework/agents/specialists/`.
- **Safety mechanisms** like the “May I write?” gate in [`.claude/statusline.sh`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/.claude/statusline.sh) prevent accidental repository modifications.
- **Transparency** is enforced by requiring visible drafts before any code or design document is finalized.

## Frequently Asked Questions

### What happens if I reject the draft during the Approval stage?

If you decline the write request, the agent returns to the Draft or Options stage to revise based on your feedback. The protocol loops back until you explicitly approve the artifact with “Yes,” ensuring the repository only contains user-validated content.

### Can an agent skip the Options stage and propose a single solution?

No. According to [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md), agents must present **2-4 concrete options** with trade-offs. Skipping this stage violates the protocol’s design principle of preventing unilateral agent decisions.

### Where is the collaborative protocol formally defined in the codebase?

The canonical definition appears in [`docs/COLLABORATIVE-DESIGN-PRINCIPLE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/COLLABORATIVE-DESIGN-PRINCIPLE.md) at lines 46-48, with a complete illustrated example spanning lines 49-90. The workflow is also referenced in [`docs/WORKFLOW-GUIDE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/docs/WORKFLOW-GUIDE.md) (lines 72-77) and summarized in the root [`CLAUDE.md`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/CLAUDE.md) file.

### Is the Approval step enforced programmatically or just by convention?

The Approval step is enforced programmatically. The repository includes hooks such as [`.claude/statusline.sh`](https://github.com/Donchitos/Claude-Code-Game-Studios/blob/main/.claude/statusline.sh) that manage the “May I write?” gate, ensuring agents cannot write to the filesystem without explicit user confirmation, eliminating accidental writes and maintaining repository security.