The Skill walks you through a 6-stage pipeline, entirely conversational — no runtime dependencies required:

• Self Mode — clone yourself from local AI conversations and writings
• Mentor Mode — clone a public figure via 6-angle parallel research (primary voice, live reactions, external views, decisions, social fragments, timeline)

Install the Skill (this is the whole installation):

mkdir -p ~/.claude/skills/digital-clone
curl -o ~/.claude/skills/digital-clone/SKILL.md \
  https://raw.githubusercontent.com/AliceLJY/digital-clone-skill/main/SKILL.md

Then in Claude Code:

帮我克隆自己 / "Clone myself from my articles and CC transcripts" 帮我克隆纳瓦尔做数字导师 / "Clone Naval as my digital mentor"

The Skill handles everything conversationally, stage by stage, with your approval at each step. All outputs go to ./clone-workspace/ in your current directory.

Requires Bun. The CLI does not run on Node.js (it uses Bun's TypeScript module resolution). If you don't use Bun, skip this section entirely — the Skill covers the full pipeline on its own.

For large corpora (thousands of transcript files), the CLI does the mechanical work faster than in-conversation processing:

git clone https://github.com/AliceLJY/digital-clone-skill.git
cd digital-clone-skill
bun install

bun run src/cli.ts init --target "Your Name" --mode self
bun run src/cli.ts ingest --source all
bun run src/cli.ts refine
bun run src/cli.ts readiness

Important: the workspace path is relative to where you run the commands. If you preprocess with the CLI, start your Claude Code session in the same directory so the Skill finds ./clone-workspace/. The refined corpus separates *-user.md (your voice — used for personality extraction) from *-assistant.md (AI replies — reference only, excluded from Soul Forging).

Privacy boundary: ingest/import applies best-effort sensitive-value redaction in memory before the first raw/*.jsonl write; it never writes an unfiltered copy and then overwrites it. Persisted raw/ files still contain private conversation text and redaction cannot recognize every credential or personal detail, so inspect them before sharing or uploading. Add --no-raw to scan and report without writing any raw corpus artifact. Refinement performs a second sanitization pass for legacy or manually supplied files. Original source files are never modified.

Set CLONE_WORKSPACE to pin the workspace to a fixed path shared between CLI and Skill sessions.

bun run src/cli.ts quality remains a compatibility alias for readiness; output and report naming use corpus readiness because these signals measure corpus sufficiency, not clone quality.

refresh can optionally pull recent memories from a RecallNest install (the author's memory system; set RECALLNEST_CLI or place it at ~/recallnest/lm). The export is captured from stdout, sanitized in memory, and only then written to refreshed/. Without RecallNest, use --skip-recallnest.

{
  "mcpServers": {
    "digital-clone": {
      "command": "bun",
      "args": ["run", "/path/to/digital-clone-skill/src/mcp-server.ts"],
      "cwd": "/path/to/digital-clone-skill"
    }
  }
}

Built by 小试AI (@AliceLJY) for the WeChat public account 我的AI小木屋.

Part of the 小试AI open-source AI workflow:

MIT