Summary

Microsoft has published mandatory requirements and best practices for bots that use AI in Teams. The OpenClaw Teams extension should implement these to provide a first-class agent experience.

What needs to change

Mandatory (per MS guidelines)

• AI-generated label: Add AIGeneratedContent entity and channelData.feedbackLoopEnabled to all outbound messages so Teams renders the "AI generated" badge
• Streaming responses (1:1 only): Use the Teams streaminfo entity protocol to progressively update messages as the LLM generates tokens, instead of waiting for the full response
• Welcome card / prompt starters: Send an Adaptive Card with example prompts when the bot is added to a conversation
• App manifest: Add copilotAgents.customEngineAgents node and set manifestVersion to devPreview

Best practices

• Feedback buttons with reflective learning: Enable Teams' native thumbs up/down UI. On negative feedback, run a background reflection where the agent reviews its response and derives learnings for future context injection
• Feedback persistence: Log feedback events to session JSONL (type: "custom", event: "feedback") for offline mining and quality analysis

Infrastructure

• Add updateActivity REST helper to the custom adapter for streaming support
• Add from field to outbound activities from the process handler context
• New config fields: welcomeCard, promptStarters, groupWelcomeCard, feedbackEnabled, feedbackReflection, feedbackReflectionCooldownMs

References

• AI UX best practices for Teams agents
• Stream bot messages REST API
• Official Teams SDK streaming implementation (packages/apps/src/http/http-stream.ts)