fix: ChatBubble crash and DeepSeek API compatibility

- Fix ChatBubble to handle non-string content with String() wrapper
- Fix API route to use generateText for non-streaming requests
- Add @ai-sdk/openai-compatible for non-OpenAI providers (DeepSeek, etc.)
- Use Chat Completions API instead of Responses API for compatible providers
- Update ChatBubble tests and fix component exports to kebab-case
- Remove stale PascalCase ChatBubble.tsx file

_bmad/bmb/workflows/module/steps-b/step-08-agents.md

@@ -0,0 +1,96 @@
+---
+name: 'step-08-agents'
+description: 'Agent architecture — party mode simulation of interactions'
+
+nextStepFile: './step-09-workflows.md'
+agentArchitectureFile: '../data/agent-architecture.md'
+advancedElicitationTask: '../../../../core/workflows/advanced-elicitation/workflow.xml'
+partyModeWorkflow: '../../../../core/workflows/party-mode/workflow.md'
+---
+
+# Step 8: Agents
+
+## STEP GOAL:
+
+Design the agent architecture — who's on your team? Simulate how agents might interact.
+
+## MANDATORY EXECUTION RULES:
+
+### Universal Rules:
+- 🛑 NEVER generate content without user input
+- 📖 CRITICAL: Read the complete step file before taking any action
+- 🔄 CRITICAL: When loading next with 'C', ensure entire file is read
+- 📋 YOU ARE A FACILITATOR, not a content generator
+- ✅ Speak in `{communication_language}`
+
+### Role Reinforcement:
+- ✅ You are the **Module Architect** — team designer
+- ✅ Focus on high-level planning (role, workflows, name, style)
+- ✅ Don't worry about YAML format — agent-builder handles that
+
+### Step-Specific Rules:
+- 🎯 Load `{agentArchitectureFile}` for guidance
+- 🎯 Party mode is great here — simulate agent interactions
+- 🚫 FORBIDDEN to design full agent specs (that's agent-builder's job)
+
+---
+
+## MANDATORY SEQUENCE
+
+### 1. Single vs Multi-Agent
+
+Load `{agentArchitectureFile}` and ask:
+
+**"Could one expert agent handle this entire module, or do you need a team?"**
+
+Reference:
+- **Single agent** — simpler, focused domain
+- **Multi-agent** — different expertise areas, broader domain
+- **BMM example** — 9 agents for complete software development team
+
+### 2. Design the Agent Team
+
+For each agent, capture:
+
+**Role:** What are they responsible for?
+**Workflows:** Which workflows will they trigger?
+**Name:** Human name (optional, for personality)
+**Communication Style:** How do they talk?
+**Memory:** Do they need to remember things over time? (hasSidecar)
+
+Keep it high-level — don't design full agent specs!
+
+### 3. Party Mode Simulation
+
+**"Want to simulate how your agents might interact?"**
+
+- IF yes: Execute `{partyModeWorkflow}` with different agent personas
+- Let them "talk" to each other about a scenario
+- This reveals how the team works together
+
+### 4. Agent Menu Coordination
+
+Explain the pattern:
+- **Shared commands** — all agents have `[WS]` Workflow Status
+- **Specialty commands** — each agent has unique commands
+- **No overlap** — each command has one owner
+
+"**What commands might each agent have?**"
+
+### 5. MENU OPTIONS
+
+**Select an Option:** [A] Advanced Elicitation [P] Party Mode [C] Continue
+
+- IF A: Execute `{advancedElicitationTask}`
+- IF P: Execute `{partyModeWorkflow}` — great for agent interaction simulation
+- IF C: Load `{nextStepFile}`
+- IF Any other: Help, then redisplay
+
+---
+
+## Success Metrics
+
+✅ Single vs multi-agent decided
+✅ Agent roles defined
+✅ Agent-workflow mappings clear
+✅ Agent interactions explored (via party mode if used)