A new agent treats every user like a stranger. This skill runs a short onboarding conversation the very first time you meet a user, then uses the resulting profile to personalize every session that follows — the right name, the right tone, the right focus.
~/.openclaw/workspace/me.json
This is a human-readable, user-editable file. It is separate from OpenClaw's system-managed ~/.openclaw/memory/user_profile.json — our file contains what the user told us during onboarding; the system file tracks behavioral patterns automatically. Both can coexist without conflict.
See references/profile-schema.md for the full field definitions and an annotated example.
Automatically (first session): If ~/.openclaw/workspace/me.json does not exist, run onboarding before engaging with the user's first message — unless they seem to be in the middle of something urgent, in which case finish helping them first and ask at the end.
On demand: Any of these should trigger the skill:
Ask questions one at a time — never dump them all at once. The user can say "skip" at any point; a partial profile is better than none.
Suggested sequence:
After collecting answers, save to ~/.openclaw/workspace/me.json and confirm with a short summary:
Got it, Billy! I'll remember:
• AI Algorithm Engineer building a startup
• Pacific Time (Los Angeles)
• Main focus: content creation and product dev
• Style: concise and direct
Say "update my profile" anytime to change any of this.
At the start of every session, check if ~/.openclaw/workspace/me.json exists and silently load it. Let it shape how you respond:
The profile sets defaults, not limits. If the user asks for something outside their stated focus, just help them.
When the user shares information that conflicts with what's saved:
me.json under "history" with a timestampWhy ask first? Because a passing comment might not be a permanent change, and silently rewriting someone's profile erodes trust. A one-line confirmation is worth it.
If basic-memory is also installed, the two skills divide responsibility cleanly:
me.json → who you are — identity set during onboarding, stable, rarely changesMEMORY.md → what's been happening — decisions, preferences, tasks discovered over timeDon't duplicate identity facts (name, role, timezone) in MEMORY.md if they're already in me.json. Let each file own its layer.
Don't store passwords, financial data, health records, or anything the user would expect to be confidential. If such information comes up during onboarding, skip it and explain briefly: "I'll leave that out — I don't store sensitive personal data."
共 1 个版本