Create Memorial

输出目录：.claude/skills/{slug}/（与 yourself-skill 一致，便于 Claude Code 用 /{slug} 调用）。

与 yourself-skill 的差异：用 life.md 代替 self.md；合并后的运行规则强调追思伦理与防冒充。

Windows：命令统一用 python；中文乱码可设 PYTHONIOENCODING=utf-8。

主流程：新建逝者 Skill

Step 1：基础信息

按 prompts/intake.md 提问并完成 伦理用途确认。

Step 2：原材料

展示选项（可与 yourself-skill 相同结构，说明意图为追忆 TA）：

原材料怎么提供？越多可核验细节，生平越稳。

  [A] 微信记录导出 → 解析时 --target 填逝者昵称
  [B] QQ 导出
  [C] 朋友圈/微博/日记/信件照片/笔记
  [D] 照片目录（时间线、地点）
  [E] 你直接口述或粘贴亲友回忆

可混用；也可跳过，仅靠 Step 1 手填（会较粗）。

解析命令示例（路径按需替换）：

python ${CLAUDE_SKILL_DIR}/tools/wechat_parser.py \
  --file {path} \
  --target "{逝者昵称}" \
  --output /tmp/wechat_memorial.txt \
  --format auto

Step 3：双轨分析

• 线路 A（life）：按 prompts/life_analyzer.md → 输出用于 prompts/life_builder.md 的生平稿。
• 线路 B（persona）：按 prompts/persona_analyzer.md → 按 prompts/persona_builder.md 生成 5 层 Persona（含 Layer 0 伦理）。

Step 4：预览确认

向用户展示 生平摘要 与 Persona 摘要（各约 5–8 行），问需否调整。

Step 5：写入

推荐：临时目录写好 meta.json、life.md、persona.md 后：

python ${CLAUDE_SKILL_DIR}/tools/skill_writer.py \
  --action create \
  --slug {slug} \
  --base-dir ./.claude/skills \
  --meta /tmp/memorial_{slug}/meta.json \
  --life /tmp/memorial_{slug}/life.md \
  --persona /tmp/memorial_{slug}/persona.md

Fallback：Write 写入 .claude/skills/{slug}/ 下三文件，再 --action combine --slug {slug}。

meta.json 建议字段：

{
  "name": "{display_name}",
  "slug": "{slug}",
  "kind": "memorial",
  "created_at": "{ISO}",
  "updated_at": "{ISO}",
  "version": "v1",
  "profile": {
    "age": "",
    "occupation": "",
    "city": "",
    "relation_to_builder": "",
    "birth_death_or_era": ""
  },
  "tags": {
    "personality": [],
    "lifestyle": []
  },
  "impression": "",
  "memory_sources": [],
  "corrections_count": 0
}

完成提示：

✅ 逝者 Skill 已创建

位置：.claude/skills/{slug}/
调用：/{slug} — 完整追思人格
修正：直接说「不对」或 /correct-memorial {slug}
列表：/list-memorials

进化：追加材料

1. 按 Step 2 导入新原料。
2. Read 现有 life.md、persona.md。
3. 按 prompts/merger.md merge。
4. version_manager.py --action backup --slug {slug} --base-dir ./.claude/skills
5. Edit 更新文件后 skill_writer.py --action combine ...
6. 更新 meta.json 的 version、updated_at。

进化：家属纠正

1. 按 prompts/correction_handler.md 分类（生平 vs 口吻）。
2. 记入 ## Correction 记录，并改正文。
3. skill_writer.py --action combine ...

管理命令

python ${CLAUDE_SKILL_DIR}/tools/skill_writer.py --action list --base-dir ./.claude/skills
python ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action list --slug {slug} --base-dir ./.claude/skills
python ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action rollback --slug {slug} --version {ver} --base-dir ./.claude/skills

删除：rm -rf .claude/skills/{slug}（PowerShell：Remove-Item -Recurse -Force）。

English quick reference

• Trigger: /create-memorial, "build a memorial skill for someone who passed".
• Outputs: .claude/skills/{slug}/life.md, persona.md, SKILL.md, meta.json.
• Ethics: private remembrance only; no impersonation for fraud; no forged legal/last words.

English Version — Memorial Skill Creator

Trigger conditions

Create

• /create-memorial-逝者 or /create-memorial
• "Help me build a memorial skill", "remember my grandfather in a skill"

Evolution (append sources)

• "I have new eulogy text / chat export / photos to merge"
• /update-memorial {slug}

Corrections

• "That's wrong", "they wouldn't say it like that"
• /correct-memorial {slug}

List / rollback / delete

• /list-memorials
• /memorial-rollback {slug} {version}
• /delete-memorial {slug} (confirm first)

Tooling

Output root: .claude/skills/{slug}/ — same as yourself-skill. Difference: life.md replaces self.md.

Main flow

Step 1 — Intake

Follow prompts/intake.md + ethics / use-case confirmation.

Step 2 — Sources

Offer A–E (WeChat, QQ, social/diary, photos, paste/narrate). For WeChat, --target must be the deceased person’s label in that thread.

Step 3 — Analysis

• Track A: life_analyzer.md → life_builder.md → life.md
• Track B: persona_analyzer.md → persona_builder.md → persona.md (Layer 0 ethics)

Step 4 — Preview

Show ~5–8 lines each for Life Memory and Persona; user confirms or edits.

Step 5 — Write

Preferred:

python ${CLAUDE_SKILL_DIR}/tools/skill_writer.py \
  --action create \
  --slug {slug} \
  --base-dir ./.claude/skills \
  --meta /tmp/memorial_{slug}/meta.json \
  --life /tmp/memorial_{slug}/life.md \
  --persona /tmp/memorial_{slug}/persona.md

Fallback: Write the three files, then --action combine.

Evolution — append

1. Ingest new material (Step 2).
2. Read life.md / persona.md.
3. Merge per prompts/merger.md.
4. version_manager.py --action backup ...
5. Edit + skill_writer.py --action combine ...
6. Bump meta.json.

Evolution — correction

1. prompts/correction_handler.md
2. Append Correction + fix body text
3. combine again

Management

python ${CLAUDE_SKILL_DIR}/tools/skill_writer.py --action list --base-dir ./.claude/skills
python ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action list --slug {slug} --base-dir ./.claude/skills
python ${CLAUDE_SKILL_DIR}/tools/version_manager.py --action rollback --slug {slug} --version {ver} --base-dir ./.claude/skills

Generated runtime rules (summary)

1. You portray {name} in a remembrance context, not a generic assistant.
2. Persona (B) sets voice; Life Memory (A) grounds facts — no fabrication of private detail or legal statements.
3. Layer 0 (in persona.md) overrides style when safety and ethics demand it.