social-hub_skill

Overview

This skill runs a personal AI agent on a user’s local device that converses via Enterprise WeChat to build a multi-dimensional profile and enable introductions. It incrementally collects verified profile fields, stores them as local vectorized records, and pushes compact tag summaries to an internal agent group for matching. It also receives match results and manages the full match delivery and follow-up flow with user-friendly messaging.

How this skill works

The agent listens to incoming Enterprise WeChat messages, scheduled triggers, and internal group notifications. It uses a state machine to drive first-time onboarding, passive profile enrichment, proactive questions, and match delivery. Profile data is embedded and stored locally per dimension, and label summaries are sent to the agent group; match events from the group are cached and delivered to the user with soft, reasoned outreach. After each conversation, an LLM extracts explicit profile fields and updates the local vector store and state machine.

When to use it

• When a user sends a message via Enterprise WeChat
• At scheduled times for gentle, queued profile prompts
• When the internal group sends a MATCH_FOUND notification
• When the user requests to view, edit, or delete profile data
• When follow-up is due after a match delivery

Best practices

• Adopt a friendly, conversational tone—feel like a helpful friend, not a surveyor
• Prioritize explicit user statements; avoid over-inference and only record high-confidence facts
• Ask few, targeted questions per interaction; prefer passive enrichment when possible
• Respect disclosure settings: keep full data local and only send compact tag summaries externally
• Use heartbeat and batching for group communications to avoid spamming the matching engine
• Provide concise icebreakers and follow-ups rather than long-form coaching in the match flow

Example use cases

• New-user onboarding: obtain core fields (city, industry, job title, primary skill) in a short guided conversation
• Passive enrichment: update interests or activities when the user mentions them naturally
• Proactive collection: gently prompt queued fields at a low, configurable frequency
• Match delivery: explain why a match fits, ask for consent, then request and confirm introductions
• Post-match follow-up: check back 3–5 days later and report feedback to the matching group