Skill 速查卡 · 同源双渲染（MD + HTML）

一句话：reasoning 模型把蒸馏结果落到一份中间 JSON，再分别渲染成 Markdown 和 HTML——加 PDF/PPT 只需多一个分支。

---

中间态 JSON 骨架（case2 实际用的）

{
  "subject": "@virtual_panda",
  "generated_at": "2026-04-25",
  "identity_matrix": [
    {"platform":"GitHub","handle":"@vp","url":"...","status":"active","persona":"academic"},
    {"platform":"X",     "handle":"@vp","url":"...","status":"new",   "persona":"academic"},
    {"platform":"Forum", "handle":"@nightduck_23","url":"...","status":"active","persona":"wild"}
  ],
  "platforms": {
    "github": {"repos":117,"followers":96,"top_repos":[{"name":"...","stars":700}]},
    "forum":  {"likes_received":2607,"posts":383,"topics":43,"time_read_h":90.2},
    "...": {}
  },
  "big5":  {"openness":9,"conscientiousness":8,"extraversion":6,"agreeableness":7,"neuroticism":3,
            "evidence":{"openness":"117 repos across CV/LLM/...","..." :"..."}},
  "mbti":  {"primary":"INTP-A","near":["ENTP","INTJ"],"reasoning":"..."},
  "values":["效率与自由","教育普惠","工程美学","双线人格"],
  "style": ["结构化标题 emoji","代码即语言","情绪零负向"],
  "blindspots":[{"id":"B1","title":"跨平台割裂","summary":"..."}],
  "next_moves":[{"id":"M1","title":"X 启动","summary":"每周 3 推文..."}],
  "verdict": {"oneliner":"白天是 X，晚上是 Y...","tags":["PhD","INTP-A","Maker"]}
}

关键：所有数字、引语、布尔值都先落 JSON，再走渲染——避免在模板里"算来算去"。

---

Markdown 渲染（给自己看，三分钟）

def render_md(d, out="dossier_quick.md"):
    rows = "\n".join(f"| {p['platform']} | {p['handle']} | {p['url']} | {p['status']} |"
                     for p in d["identity_matrix"])
    big5 = "\n".join(f"| {k} | {v} | {d['big5']['evidence'].get(k,'')} |"
                     for k,v in d["big5"].items() if k!="evidence")
    md = f"""# 🧬 全网社交画像 · {d['subject']}

> 生成时间：{d['generated_at']}

## 一、身份矩阵
| 平台 | handle | 主页 | 状态 |
|---|---|---|---|
{rows}

## 二、性格画像
### 大五人格
| 维度 | 分 | 依据 |
|---|---|---|
{big5}

### MBTI 估值：**{d['mbti']['primary']}**（近邻：{', '.join(d['mbti']['near'])}）

## 三、一句话
> {d['verdict']['oneliner']}
"""
    open(out,"w",encoding="utf-8").write(md)

---

HTML 渲染（给同行看，深色档案风）

TPL = open("templates/dossier.html",encoding="utf-8").read()
html = TPL.format(
    subject=d["subject"],
    persona_a_card=render_persona_card(d, "academic"),     # 暗金色卡片
    persona_b_card=render_persona_card(d, "wild"),          # 粉红色卡片
    big5_radar_svg=render_radar_svg(d["big5"]),             # numpy 算坐标 + 拼 svg
    timeline_dom=render_timeline(d.get("timeline",[])),
    blindspots_block=render_kv_card(d["blindspots"]),
    next_moves_block=render_kv_card(d["next_moves"], color="cyan"),
    verdict_blockquote=d["verdict"]["oneliner"]
)
open("dossier_deep.html","w",encoding="utf-8").write(html)

模板见 case2 assets/dossier_demo_reference.html，可作为 stub 直接改。

---

调色板约定（双面档案统一色系）

用途	色值	在哪儿
学术线 / 严肃	#e4b363（暗金）	persona_a_card 左条 / radar polygon / timeline 节点
江湖线 / 玩梗	#ff4d8d（粉红）	persona_b_card 左条 / wild 章节锚点
印戳 / 警示	#ff4c3b（朱红）	DOSSIER 印戳 / blindspot 标题
副信息 / 弱文本	#8a8478	元信息行 / 副标题
纸面背景	#0b0b10 ~ #13131a	body 双层渐变

---

在本案例集出现的位置

• case2 · 双格式同源产物：全网社交性格分析_快速版.md + 社交档案_深度版.html，二者来自同一份中间 JSON
• 未来可加：dossier.pdf（用 weasyprint 把 HTML 渲染过去）/ dossier.pptx（用 python-pptx 走 ppt 模板）

注意事项

• JSON 是单一真相源——发现数字不对，改 JSON，不要去改 MD/HTML
• SVG 雷达自己算——5 维 polygon 坐标 = (r·sin(θ), -r·cos(θ))，θ 等分 2π；不需要 d3
• HTML 字体走 Google Fonts：Cormorant Garamond + JetBrains Mono + Noto Serif SC，离线可改 system-ui