diff --git a/README.md b/README.md
index ac8eec4..f9dd897 100644
--- a/README.md
+++ b/README.md
@@ -35,6 +35,12 @@ The Skill walks you through a 6-stage pipeline, entirely conversational — no r
 - **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)
 
+### X/Twitter Source Packets
+
+For Mentor Mode, X/Twitter posts can strengthen the Social Fragments angle when they are treated as evidence, not personality conclusions. If a user needs a reproducible public corpus, [TweetClaw](https://github.com/Xquik-dev/tweetclaw) can optionally collect public tweets, replies, media context, source URLs, and capture timestamps for review before Stage 2 ingestion.
+
+Keep Digital Clone responsible for evidence grading, refinement, negative-space analysis, and verification. Use only public material or consented exports, and do not ingest private DMs, cookies, private-account material, or non-consented personal archives.
+
 ---
 
 ## Quick Start
@@ -158,7 +164,7 @@ Part of the **小试AI** open-source AI workflow:
 | Project | Description |
 |---------|-------------|
 | [recallnest](https://github.com/AliceLJY/recallnest) | MCP memory workbench (LanceDB + Jina v5) |
-| [content-publisher](https://github.com/AliceLJY/content-publisher) | Image generation + layout + WeChat publishing |
+| content-publisher | Image generation + layout + WeChat publishing |
 | [openclaw-tunnel](https://github.com/AliceLJY/openclaw-tunnel) | Docker ↔ host CLI bridge (/cc /codex /gemini) |
 | [telegram-ai-bridge](https://github.com/AliceLJY/telegram-ai-bridge) | Telegram bots for Claude, Codex, and Gemini |
 | [claude-code-studio](https://github.com/AliceLJY/claude-code-studio) | Multi-session collaboration platform for Claude Code |
diff --git a/SKILL.md b/SKILL.md
index 21303a8..b1e81d2 100644
--- a/SKILL.md
+++ b/SKILL.md
@@ -142,6 +142,13 @@ Launch 6 parallel sub-agents, each targeting a different **analysis angle**. Dif
 | Others' analysis | Blind spots + external patterns | Medium |
 | Secondhand accounts | Reference only, needs verification | Low |
 
+**Optional X/Twitter source packets:**
+- Use only public posts, public replies, public media context, or consented exports.
+- If the user needs reproducible collection, [TweetClaw](https://github.com/Xquik-dev/tweetclaw) can collect source material from OpenClaw or npm before Stage 2 ingestion.
+- Save the packet to `references/research/05-social-fragments.md` or `raw/social-fragments/` with `source_url`, `captured_at`, `query`, `author_handle`, `post_text_excerpt`, `media_context`, `reply_context`, and `credibility`.
+- Treat TweetClaw output as source evidence only. Digital Clone still grades credibility, refines corpus, identifies negative space, and verifies the clone.
+- Do not ingest private DMs, cookies, private-account material, or non-consented personal archives.
+
 **Scaling rule:** For info-sparse targets (< 10 findable sources), reduce to 3 agents (Primary Voice + Live Reactions + External Views), skip the rest. Flag to user: "信息有限，克隆精度会降低。"
 
 **Core principle (from nuwa-skill):** 宁可生成一个诚实的 60 分克隆并标注清楚局限，也不要一个看似完美但编造了信息的 90 分克隆。
@@ -209,6 +216,7 @@ Stage 1 agents already completed the main research. Stage 2 focuses on **convert
    ```
    from:[username] until:[end-date] -filter:retweets
    ```
+   For reproducible social fragments, ask the user to export or collect a source packet that keeps URLs, capture timestamps, queries, and short excerpts together. TweetClaw is an optional collector for public X/Twitter material; it does not decide clone personality or evidence strength.
 
    **Podcast/video transcripts:**
    - Check if text transcripts exist first