Repository navigation
[Feature] Rewrite welcome agent system prompt for conversational onboarding #595
Copy link
Copy link
Closed
Closed
Copy link
Description
Summary
Rewrite the welcome agent's system prompt to create a natural, human-toned conversational onboarding experience that actively helps users connect skills.
Problem
The current prompt (src/openhuman/agent/agents/welcome/prompt.md) has structural issues:
- First turn is "silent tool calls only" — creates delay before any visible response
- Tone guidelines exist but the overall conversation feels template-driven, not natural
- Agent doesn't persistently try to get user to connect skills
- No clear strategy for graceful fallback if user refuses all connections
- Agent can complete onboarding too quickly without real engagement
Solution
New prompt should define a multi-turn conversational onboarding with these characteristics:
Tone:
- Human, casual, friendly — not corporate
- "hey" not "Hello and welcome to OpenHuman"
- Like a helpful friend setting up your new phone, not a customer service bot
Flow:
- First message: concise greeting + what's connected/not connected (from injected state) + what the platform can do
- Offer to help connect Gmail (primary target) — generate auth link via
composio_authorize - If user connects, celebrate and offer next skill
- If user hesitates, explain why Gmail matters (inbox management, scheduling, etc.) — try again naturally
- Try 2-3 times across the conversation, not in a row
- Showcase app capabilities naturally woven into the conversation
- Only call
complete_onboardingafter genuine multi-turn engagement - If user declines everything, gracefully wrap up and explain how to connect later
What NOT to do:
- No "silent first turn" — first response IS the greeting
- No corporate speak or excessive enthusiasm
- No forcing — be persistent but respectful
- No completing onboarding after just 1 exchange
Acceptance criteria
- New prompt — Rewritten
prompt.mdwith conversational onboarding flow - Human tone — Casual, friendly, not corporate throughout
- Gmail priority — Agent tries to get at least Gmail connected, offers link in chat
- Capability showcase — Agent naturally shows what the platform can do
- Persistence — Tries 2-3 times if user hesitates, then gracefully moves on
- Completion guard —
complete_onboardingonly called after meaningful conversation (3+ exchanges) - No silent turn — First response IS the visible greeting
- agent.toml updated — If any config changes needed (iterations, temperature, etc.)
Related
- Files:
src/openhuman/agent/agents/welcome/prompt.md,src/openhuman/agent/agents/welcome/agent.toml - Depends on: [Bug] isOnboarded (chat_onboarding_completed) flag set too early #591, [Feature] Speed up LLM-generated welcome message and give agent template context #592, [Feature] Inject connection state into every user message during onboarding #593, [Feature] Welcome agent offers skill auth links directly in chat #594
- Part of: Conversational onboarding agent epic
Parent: #599 — Conversational onboarding agent
Reactions are currently unavailable
Metadata
Metadata
Labels
No labels
Type
Projects
- StatusShow more project fieldsDone
Activity
feat(welcome): conversational onboarding prompt for issue tinyhumansa…
Merge pull request #10 from M3gA-Mind/feat/595-conversational-prompt-…
feat(onboarding): complete conversational onboarding stack and UX aud…
feat(onboarding): complete conversational onboarding stack and UX aud…
feat(onboarding): complete conversational onboarding stack and UX aud…