Copy My Profile

Extract a standardized cross-tool user profile from any AI agent's long-term memory (USER.md, MEMORY.md, AGENTS.md, CLAUDE.md, .cursorrules, etc.) so you can re-use it in another AI tool without retraining. Output is a portable Markdown document with five categories — instructions, identity, profession, projects, preferences. Works with Claude Code, OpenClaw, Codex CLI, Cursor, Cline, Continue, Aider, or any LLM agent. 适用场景:生成我的画像/导出 profile/ 拷贝到其他 AI 工具/跨工具同步偏好/换 agent 不丢记忆。

• Version: 1.0.0
• License: MIT
• Author: Evan Song · github.com/Songhonglei
• Repository: https://github.com/Songhonglei/better-agent-skills

Generate a portable Markdown profile from one AI tool's long-term memory and re-use it in another. Solves the "I just trained Tool A to know me, now I want to use Tool B" problem without retraining.

Why this skill

Modern developers use multiple AI tools: Claude Code at home, Cursor at work, ChatGPT for brainstorming, OpenClaw for backend automation, etc. Each tool accumulates its own memory of who you are — but none of them talk to each other. This skill defines a portable Markdown format (like vCard for contacts, or ICS for calendars) so your "AI profile" can move with you.

• Zero infrastructure: no cloud, no account, no sync server
• Privacy-controlled: you see the full output before copying anywhere
• Standard format: 5 categories that any LLM can parse reliably

Output categories (5 in order)

Detailed format examples and filtering rules live in references/profile-template.md (read it during step 2-3).

1. Instructions — Explicit rules the user wants followed (tone, format, behavior corrections, ⚠️-marked rules in memory files)
2. Identity — Name, education, interests, non-sensitive personal info
3. Profession — Role, company, skill domains
4. Projects — Projects the user actually built or invested effort in
5. Preferences — Broadly-applicable work style and taste preferences

Execution steps

Step 1 — Pick execution path

Choose based on the current runtime environment.

Path A — Memory-search-capable environment (Claude Code, OpenClaw, Codex CLI, or any agent that exposes a memory_search / file-read tool):

1. Run memory_search for topics like "instructions", "preferences", "projects"
2. Read key files based on the host tool:
   • Claude Code: ~/.claude/CLAUDE.md, project CLAUDE.md
   • OpenClaw: USER.md, MEMORY.md, AGENTS.md, memory/*.md
   • Codex CLI: AGENTS.md, ~/.codex/instructions.md
   • Cursor: project .cursorrules, .cursor/rules/*.md
   • Cline: project .clinerules
   • Continue: ~/.continue/config.json (system message), workspace .continue/*.md
   • Aider: CONVENTIONS.md, .aider.conf.yml
   • See references/tools-memory-map.md for the full table.
3. Read the most recent N daily memory logs (if the tool keeps them) to capture recent context

Path B — No memory-search tool (Cline without filesystem, web ChatGPT, fresh Cursor session, etc.):

1. Ask the user to provide files or paste relevant content (see references/tools-memory-map.md for what to ask for)
2. If the user only supplies an old profile, normalize the format and note at the end: "Based on existing profile, not re-extracted from raw memory"

Degradation strategy (both paths):

• If a key file is missing, continue with what's available and list "missing sources" at the bottom
• If a category has zero data, output (no data) — don't skip the category
• If memory_search returns empty, fall back to Path B

Step 2 — Filter

• Instructions: Only items clearly identifiable as rules in memory files; don't fabricate. Distinction rules in profile-template.md
• Identity: Only non-sensitive info the user has actively shared. Privacy filter rules in profile-template.md
• Projects: One line per project — function, status, key decisions
• Preserve the user's original phrasing as much as possible

Step 3 — Output format

Write the profile to ./my-profile.md (current directory) and also reply to the user inline. Format:

• Each line: [YYYY-MM-DD] - entry content
• Use [unknown] if the date is unclear
• Within each category, sort by date ascending
• Wrap the entire profile in a single fenced Markdown code block (so the user can copy-paste in one click)

Step 4 — Closing notes

After the code block, add a brief note covering:

• Whether all relevant info from current memory was included
• Any categories that had no data or used degradation
• Any dimensions or uncertain entries excluded (so the user can decide whether to add them)
• Import hint: which file to paste this into for the target tool — see references/import-prompts.md for ready-to-use prompts

Tips for cross-tool transfer

• Pull (export): Run this skill in your source tool → copy my-profile.md
• Push (import): In the target tool, paste the profile with a prompt from references/import-prompts.md
• The Markdown is plain text — works through clipboard, email, Notion, GitHub gist, file transfer, anything

See also

• references/profile-template.md — full output format example + filtering rules + privacy rules
• references/tools-memory-map.md — where each AI tool stores its memory (where to read FROM and where to write TO)
• references/import-prompts.md — copy-paste prompts for importing the profile into each target tool