@shulex/cxclawctl 0.0.0-stage → 0.1.2

package/skills/poc/scripts/gen_excel.py

@@ -0,0 +1,234 @@
+"""
+Generate intent classification detail Excel from items_*.json + results_*.json.
+
+Columns: L1 Intent | L2 Intent | Case Text
+L1 and L2 cells are merged across rows of the same value.
+
+Usage:
+  python3 scripts/gen_excel.py [--brand "Acme"]
+"""
+
+import os, sys
+sys.path.insert(0, os.path.join(os.environ.get("SOLVEA_LIB_DIR", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "__lib__"))), "skill-runtime"))
+from bootstrap import *
+
+import json, argparse
+import openpyxl
+
+
+def _read_state() -> dict:
+    _sf = STATE_DIR / "progress.json"
+    if _sf.exists():
+        try:
+            return json.loads(_sf.read_text(encoding="utf-8"))
+        except Exception:
+            pass
+    return {}
+
+
+def _write_state(data: dict) -> None:
+    _sf = STATE_DIR / "progress.json"
+    _sf.parent.mkdir(parents=True, exist_ok=True)
+    _sf.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+from openpyxl.styles import PatternFill, Font, Alignment, Border, Side
+from openpyxl.utils import get_column_letter
+
+
+COLOR_PALETTE = [
+    "FEF3C7", "DBEAFE", "D1FAE5", "FCE7F3", "EDE9FE",
+    "FEE2E2", "ECFDF5", "FFF7ED", "F0F9FF", "F7FEE7",
+]
+
+
+def thin_border():
+    s = Side(style="thin", color="D1D5DB")
+    return Border(left=s, right=s, top=s, bottom=s)
+
+
+def build_excel(items, results, brand=""):
+    wb = openpyxl.Workbook()
+    ws = wb.active
+    ws.title = "Intent Detail"
+
+    all_intents = results.get("intents", []) + results.get("new_intents", [])
+    l1_color = {}
+    for i, intent in enumerate(all_intents):
+        l1_color[intent["name"]] = COLOR_PALETTE[i % len(COLOR_PALETTE)]
+
+    l1_canonical_l2s = {}
+    for intent in all_intents:
+        l1_canonical_l2s[intent["name"]] = {
+            s["name"] for s in intent.get("sub_intents", []) if s["name"] != "Other"
+        }
+
+    def canonical_l2(l1, l2):
+        allowed = l1_canonical_l2s.get(l1, set())
+        if not allowed:
+            return l2 or "Other"
+        return l2 if l2 in allowed else "Other"
+
+    l1_order = {intent["name"]: i for i, intent in enumerate(all_intents)}
+    items_sorted = sorted(items, key=lambda x: (
+        l1_order.get(x["l1"], 999),
+        canonical_l2(x["l1"], x.get("l2") or ""),
+    ))
+
+    headers = ["L1意图", "L2意图", "分类依据(LLM)", "title", "first_message", "messages"]
+    col_widths = [22, 28, 45, 25, 55, 70]
+
+    header_fill = PatternFill("solid", fgColor="1E3A5F")
+    header_font = Font(bold=True, color="FFFFFF", size=11)
+    header_align = Alignment(horizontal="center", vertical="center", wrap_text=True)
+
+    for col, (h, w) in enumerate(zip(headers, col_widths), start=1):
+        cell = ws.cell(row=1, column=col, value=h)
+        cell.fill = header_fill
+        cell.font = header_font
+        cell.alignment = header_align
+        cell.border = thin_border()
+        ws.column_dimensions[get_column_letter(col)].width = w
+
+    ws.row_dimensions[1].height = 30
+    ws.freeze_panes = "A2"
+
+    row_num = 2
+
+    for item in items_sorted:
+        l1 = item["l1"]
+        l2 = item.get("l2") or "(unclassified)"
+        title_val = item.get("title", "")
+        first_msg_val = item.get("first_message", "") or item.get("text", "")
+        messages_val = item.get("messages", "")
+        raw_text = item.get("text", "")
+        text_val = raw_text[:1000] + f"...[截断 {len(raw_text)-1000} 字]" if len(raw_text) > 1000 else raw_text
+        color = l1_color.get(l1, "F9FAFB")
+
+        fill = PatternFill("solid", fgColor=color)
+        border = thin_border()
+
+        c1 = ws.cell(row=row_num, column=1, value=l1)
+        c1.fill = fill
+        c1.border = border
+        c1.alignment = Alignment(horizontal="center", vertical="top", wrap_text=True)
+        c1.font = Font(bold=True, size=10)
+
+        l2_display = canonical_l2(l1, l2)
+        l2_label = l2_display
+        if item.get("is_new_l2") and l2_display not in ("Other", ""):
+            l2_label = f"★ {l2_display}"
+        if item.get("is_new_l1"):
+            l2_label = f"[新] {l2_display}"
+
+        c2 = ws.cell(row=row_num, column=2, value=l2_label)
+        c2.fill = PatternFill("solid", fgColor="F8FAFC")
+        c2.border = border
+        c2.alignment = Alignment(horizontal="left", vertical="top", wrap_text=True)
+        c2.font = Font(size=10)
+
+        for col_idx, val in [(3, text_val), (4, title_val), (5, first_msg_val), (6, messages_val)]:
+            cx = ws.cell(row=row_num, column=col_idx, value=val)
+            cx.fill = PatternFill("solid", fgColor="FFFFFF")
+            cx.border = border
+            cx.alignment = Alignment(horizontal="left", vertical="top", wrap_text=True)
+            cx.font = Font(size=9)
+        ws.row_dimensions[row_num].height = max(15, min(80, len(first_msg_val) // 3))
+
+        row_num += 1
+
+    # Summary sheet
+    ws2 = wb.create_sheet("Summary")
+    ws2.column_dimensions["A"].width = 28
+    ws2.column_dimensions["B"].width = 28
+    ws2.column_dimensions["C"].width = 10
+    ws2.column_dimensions["D"].width = 10
+
+    h_fill = PatternFill("solid", fgColor="1E3A5F")
+    h_font = Font(bold=True, color="FFFFFF", size=11)
+    for col, val in enumerate(["一级意图", "二级意图", "数量", "占比 %"], start=1):
+        c = ws2.cell(row=1, column=col, value=val)
+        c.fill = h_fill
+        c.font = h_font
+        c.alignment = Alignment(horizontal="center", vertical="center")
+        c.border = thin_border()
+    ws2.row_dimensions[1].height = 28
+    ws2.freeze_panes = "A2"
+
+    sr = 2
+    for intent in all_intents:
+        l1n = intent["name"]
+        color = l1_color.get(l1n, "F9FAFB")
+        fill = PatternFill("solid", fgColor=color)
+        subs = intent.get("sub_intents", [])
+        if not subs:
+            c = ws2.cell(row=sr, column=1, value=l1n)
+            c.fill = fill; c.font = Font(bold=True, size=10); c.border = thin_border()
+            c.alignment = Alignment(horizontal="center", vertical="center", wrap_text=True)
+            ws2.cell(row=sr, column=2, value="(no sub-intents)").border = thin_border()
+            ws2.cell(row=sr, column=3, value=intent["count"]).border = thin_border()
+            ws2.cell(row=sr, column=4, value=round(intent["pct"], 1)).border = thin_border()
+            sr += 1
+        else:
+            l1_start_row = sr
+            for sub in subs:
+                c = ws2.cell(row=sr, column=1, value=l1n)
+                c.fill = fill; c.font = Font(bold=True, size=10); c.border = thin_border()
+                c.alignment = Alignment(horizontal="center", vertical="center", wrap_text=True)
+                ws2.cell(row=sr, column=2, value=sub["name"]).border = thin_border()
+                ws2.cell(row=sr, column=2).alignment = Alignment(horizontal="left", vertical="center")
+                ws2.cell(row=sr, column=3, value=sub["count"]).border = thin_border()
+                ws2.cell(row=sr, column=3).alignment = Alignment(horizontal="center")
+                ws2.cell(row=sr, column=4, value=round(sub["pct"], 1)).border = thin_border()
+                ws2.cell(row=sr, column=4).alignment = Alignment(horizontal="center")
+                sr += 1
+            if sr - 1 > l1_start_row:
+                ws2.merge_cells(start_row=l1_start_row, start_column=1,
+                                end_row=sr - 1, end_column=1)
+                ws2.cell(row=l1_start_row, column=1).alignment = Alignment(
+                    horizontal="center", vertical="center", wrap_text=True)
+
+    return wb
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--brand", default="")
+    parser.add_argument("--results", default=None,
+                        help="Override results JSON path (default: read from state)")
+    parser.add_argument("--items", default=None,
+                        help="Override items JSON path (default: read from state)")
+    args = parser.parse_args()
+
+    state = _read_state()
+    results_path = args.results or state.get("results_path")
+    items_path = args.items or state.get("items_path")
+
+    if not results_path or not os.path.exists(results_path):
+        raise FileNotFoundError(
+            f"Results file missing (state.results_path={results_path}). Run classify.py first."
+        )
+    if not items_path or not os.path.exists(items_path):
+        raise FileNotFoundError(
+            f"Items file missing (state.items_path={items_path}). "
+            "Run classify.py with --xlsx first."
+        )
+
+    with open(items_path) as f:
+        items = json.load(f)
+    with open(results_path) as f:
+        results = json.load(f)
+
+    wb = build_excel(items, results, args.brand)
+
+    OUTPUTS_DIR.mkdir(parents=True, exist_ok=True)
+    out_path = str(OUTPUTS_DIR / f"detail_{timestamp()}.xlsx")
+    wb.save(out_path)
+    print(f"[gen_excel] Excel -> {out_path} ({len(items)} rows)")
+
+    state = _read_state()
+    state["excel_path"] = out_path
+    state["updated_at"] = timestamp()
+    _write_state(state)
+
+
+if __name__ == "__main__":
+    main()

package/skills/poc/scripts/gen_poc_manual.py

@@ -0,0 +1,249 @@
+#!/usr/bin/env python3
+"""POC 测试手册 HTML 生成器。
+
+读取 poc_test.py 输出的 results JSON + 一份模板 + 一份 CSS，
+渲染成单文件 HTML（图片 base64 嵌入，无外部依赖、可邮件直发）。
+
+模板路由：
+  .html  → Jinja2 渲染（推荐）；py 侧传 by_agent dict，模板自己写循环
+  .md    → 传统 markdown 渲染（旧模板，保留供回退）
+
+Jinja2 模板可用变量：
+  顶层：title, brand_name, brand_website, generated_at, total_cases,
+        console_url, agents_url, account_email, account_password,
+        scene_count, by_agent
+  by_agent：dict[str, list[case]]，已过滤掉无 AI 回复的 case 和空 agent，
+            case 字段：name, agent_name, agent_id, scenario, results, stopped_reason
+            results 字段：turn, input, content, chatId, handoff, ...
+
+旧 .md 模板额外占位符（{{var}} 替换）：
+  scene_tags_html, agent_sections_html
+
+用法：
+  python3 gen_poc_manual.py --results results.json \\
+      --output manual.html \\
+      [--template assets/poc_test_template.html] \\
+      [--css assets/default_manual.css] \\
+      [--brand-name WOLTU] [--brand-website https://woltu.de] \\
+      [--account-email a@b.com] [--password 123456] [--title "POC 测试手册"]
+"""
+
+import argparse
+import base64
+import json
+import mimetypes
+import os
+import re
+import sys
+from collections import defaultdict
+from datetime import datetime
+from pathlib import Path
+
+sys.path.insert(0, os.path.join(os.environ.get("SOLVEA_LIB_DIR", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "__lib__"))), "skill-runtime"))
+from bootstrap import *  # noqa: F401,F403
+
+import markdown  # noqa: E402
+from jinja2 import Environment, BaseLoader  # noqa: E402
+
+ASSETS_DIR = Path(__file__).resolve().parent.parent / "assets"
+DEFAULT_CONSOLE_URL = "https://apps.voc.ai/chatbot#/setting"
+DEFAULT_AGENTS_URL = "https://apps.voc.ai/chatbot#/agents"
+DEFAULT_TEMPLATE = str(ASSETS_DIR / "poc_test_template.html")
+DEFAULT_CSS = str(ASSETS_DIR / "default_manual.css")
+
+
+# ── Helpers ──────────────────────────────────────────────────────────────────
+
+def _esc(s: str) -> str:
+    return (s or "").replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
+
+
+def _has_reply(case: dict) -> bool:
+    """case 中是否有至少一轮 AI 实际回复了内容（过滤掉全部出错 / 无 content 的 case）。"""
+    return any(t.get("content") for t in case.get("results", []))
+
+
+# ── MD-path rendering (legacy, kept for rollback) ────────────────────────────
+
+def _render_dialogues(turns) -> str:
+    rows = []
+    for t in turns:
+        u = _esc(t.get("input", ""))
+        b = _esc(t.get("content", "")).replace("\n", "<br>")
+        rows.append(
+            f'<div class="msg"><span class="label">👤 User</span>'
+            f'<div class="bubble user-bubble">{u}</div></div>'
+        )
+        rows.append(
+            f'<div class="msg"><span class="label">🤖 AI</span>'
+            f'<div class="bubble bot-bubble">{b}</div></div>'
+        )
+    return "\n".join(rows)
+
+
+def _render_cases(cases) -> str:
+    out = []
+    for c in cases:
+        results = c.get("results", [])
+        chat_id = results[-1].get("chatId", "—") if results else "—"
+        scenario = c.get("scenario") or c.get("name", "")
+        case_id = c.get("name", "")
+        out.append(f"""<div class="case-card">
+<div class="case-header">
+  <span class="case-id">{_esc(case_id)}</span>
+  <span class="case-scenario">{_esc(scenario)}</span>
+  <span class="chat-id-badge">Chat ID: <code>{_esc(chat_id)}</code></span>
+</div>
+<div class="dialogue">{_render_dialogues(results)}</div>
+</div>""")
+    return "\n".join(out)
+
+
+def _render_agent_sections(by_agent: dict) -> str:
+    out = []
+    for aname, cases in by_agent.items():
+        visible = [c for c in cases if _has_reply(c)]
+        if not visible:
+            continue
+        aid = visible[0].get("agent_id", "")
+        out.append(
+            f'<section class="agent-section">'
+            f'<h2>🤖 {_esc(aname)} <span class="agent-meta">Agent ID: {aid}</span></h2>'
+            f'{_render_cases(visible)}</section>'
+        )
+    return "\n".join(out)
+
+
+def _render_scene_tags(by_agent: dict) -> str:
+    return "".join(f'<span class="scene-tag">{_esc(n)}</span>' for n in by_agent.keys())
+
+
+def render_template(tpl: str, ctx: dict) -> str:
+    """轻量模板替换：仅 {{var}} 占位符（旧 .md 路径使用）。"""
+    def _sub(match):
+        key = match.group(1).strip()
+        val = ctx.get(key, "")
+        return str(val) if val is not None else ""
+    return re.sub(r"\{\{\s*([\w\.]+)\s*\}\}", _sub, tpl)
+
+
+# ── Image embedding ───────────────────────────────────────────────────────────
+
+def _embed_images_as_base64(html: str, base_dir: Path) -> str:
+    """把 <img src="<rel_path>"> 替换成 data: base64 内嵌。
+
+    跳过远程 URL（http/https/data:）。本地路径找不到时打 warning 但不阻断渲染。
+    """
+    def _sub(m):
+        prefix = m.group(1)
+        src = m.group(2)
+        suffix = m.group(3)
+        if src.startswith(("http://", "https://", "data:")):
+            return m.group(0)
+        img_path = (base_dir / src).resolve()
+        if not img_path.exists():
+            print(f"[warn] image not found, kept original src: {img_path}", file=sys.stderr)
+            return m.group(0)
+        mime, _ = mimetypes.guess_type(str(img_path))
+        if not mime:
+            mime = "image/png"
+        b64 = base64.b64encode(img_path.read_bytes()).decode("ascii")
+        return f'{prefix}data:{mime};base64,{b64}{suffix}'
+    return re.sub(r'(<img[^>]*\ssrc=")([^"]+)(")', _sub, html)
+
+
+def _wrap_html(body: str, title: str, css: str) -> str:
+    return f"""<!doctype html>
+<html lang="zh-CN">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>{_esc(title)}</title>
+<style>
+{css}
+</style>
+</head>
+<body>
+{body}
+</body>
+</html>
+"""
+
+
+# ── Main ──────────────────────────────────────────────────────────────────────
+
+def main():
+    parser = argparse.ArgumentParser(description="POC 测试手册 HTML 生成器")
+    parser.add_argument("--results", required=True, help="poc_test.py 输出的 results JSON 路径")
+    parser.add_argument("--output", required=True, help="输出 HTML 路径")
+    parser.add_argument("--title", default="Solvea POC 测试手册", help="手册标题")
+    parser.add_argument("--brand-name", default="", help="客户品牌名")
+    parser.add_argument("--brand-website", default="", help="客户品牌官网 URL")
+    parser.add_argument("--console-url", default=DEFAULT_CONSOLE_URL)
+    parser.add_argument("--agents-url", default=DEFAULT_AGENTS_URL)
+    parser.add_argument("--account-email", default=os.environ.get("SOLVEA_ACCOUNT_EMAIL", "请联系实施顾问获取"))
+    parser.add_argument("--password", default="123456")
+    args = parser.parse_args()
+
+    summary = json.loads(Path(args.results).read_text(encoding="utf-8"))
+    results = summary.get("results", [])
+
+    # 按 agent 分组，并过滤掉无 AI 回复的 case 和全空的 agent section
+    by_agent_raw: dict = defaultdict(list)
+    for r in results:
+        by_agent_raw[r.get("agent_name") or "(unknown)"].append(r)
+
+    by_agent = {
+        aname: visible
+        for aname, cases in by_agent_raw.items()
+        if (visible := [c for c in cases if _has_reply(c)])
+    }
+
+    tpl_path = Path(DEFAULT_TEMPLATE)
+    tpl = tpl_path.read_text(encoding="utf-8")
+
+    ctx = {
+        "title": args.title,
+        "brand_name": args.brand_name,
+        "brand_website": args.brand_website,
+        "generated_at": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
+        "total_cases": len(results),
+        "console_url": args.console_url,
+        "agents_url": args.agents_url,
+        "account_email": args.account_email,
+        "account_password": args.password,
+        "scene_count": len(by_agent),
+        "by_agent": by_agent,
+    }
+
+    if tpl_path.suffix.lower() == ".html":
+        # Jinja2 渲染：模板自带 CSS + 循环，直接输出完整 HTML
+        env = Environment(loader=BaseLoader(), autoescape=False)
+        template = env.from_string(tpl)
+        html = template.render(**ctx)
+        html = _embed_images_as_base64(html, base_dir=tpl_path.parent)
+    else:
+        # 旧 .md 路径（保留供回退）
+        md_ctx = {
+            **ctx,
+            "scene_tags_html": _render_scene_tags(by_agent),
+            "agent_sections_html": _render_agent_sections(by_agent_raw),
+        }
+        rendered = render_template(tpl, md_ctx)
+        body = markdown.markdown(
+            rendered,
+            extensions=["tables", "fenced_code", "md_in_html", "sane_lists"],
+        )
+        body = _embed_images_as_base64(body, base_dir=tpl_path.parent)
+        css = Path(args.css).read_text(encoding="utf-8")
+        html = _wrap_html(body, title=args.title, css=css)
+
+    out_path = Path(args.output)
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    out_path.write_text(html, encoding="utf-8")
+    size_kb = out_path.stat().st_size / 1024
+    print(f"  Manual rendered → {out_path}  ({size_kb:.1f} KB)", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    main()

package/skills/poc/scripts/gen_report.py

@@ -0,0 +1,208 @@
+"""
+HTML report generator for poc skill.
+
+Reads classification results (workspace/state/poc/progress.json -> results_path) and
+renders a polished HTML report under workspace/outputs/poc/.
+
+Usage:
+  python3 scripts/gen_report.py [--brand "Acme"]
+"""
+
+import os, sys
+sys.path.insert(0, os.path.join(os.environ.get("SOLVEA_LIB_DIR", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "__lib__"))), "skill-runtime"))
+from bootstrap import *
+
+import json, argparse
+from uuid import uuid4
+
+
+def _read_state() -> dict:
+    _sf = STATE_DIR / "progress.json"
+    if _sf.exists():
+        try:
+            return json.loads(_sf.read_text(encoding="utf-8"))
+        except Exception:
+            pass
+    return {}
+
+
+def _write_state(data: dict) -> None:
+    _sf = STATE_DIR / "progress.json"
+    _sf.parent.mkdir(parents=True, exist_ok=True)
+    _sf.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+
+from render import render_html
+
+COLOR_PALETTE = [
+    "#ef4444", "#f97316", "#eab308", "#22c55e",
+    "#14b8a6", "#3b82f6", "#8b5cf6", "#ec4899",
+    "#06b6d4", "#84cc16",
+]
+
+CIRCLE_NUMS = "①②③④⑤⑥⑦⑧⑨⑩"
+
+
+def build_config_from_results(results, brand_name=""):
+    def uid():
+        return uuid4().hex
+
+    agents = []
+    all_intents = results.get("intents", []) + results.get("new_intents", [])
+    meta = results.get("meta", {})
+    total = meta.get("total", 0)
+
+    for i, intent in enumerate(all_intents):
+        sub_intents = intent.get("sub_intents", [])
+        is_new = intent.get("is_new", False)
+        agent_id = f"intent-{i}"
+
+        l2_nodes = []
+        for j, s in enumerate(sub_intents):
+            sample_preview = s["samples"][0][:50] + "..." if s.get("samples") else ""
+            l2_nodes.append({
+                "id": f"n2-{uid()[:8]}",
+                "name": s["name"],
+                "desc": f"{s['pct']:.1f}% — {sample_preview}" if sample_preview else f"{s['pct']:.1f}%",
+                "count": s["count"],
+                "definition": s.get("definition", ""),
+                "is_shared": False,
+                "shared_agents": [],
+                "samples": s.get("samples", []),
+            })
+
+        nodes_ac = []
+        for s in sub_intents[:3]:
+            nodes_ac.append({
+                "name": f"节点 · {s['name']}",
+                "trigger": f"客户投诉：{s['name']}",
+                "action": f"<code>LLM</code> 处理 {s['name']}，引导解决",
+            })
+
+        agents.append({
+            "id": agent_id,
+            "name": f"{CIRCLE_NUMS[i % len(CIRCLE_NUMS)]} {intent['name']}",
+            "color": COLOR_PALETTE[i % len(COLOR_PALETTE)],
+            "l1_intents": [intent["name"]],
+            "total_count": intent["count"],
+            "total_pct": intent["pct"],
+            "is_new": is_new,
+            "intent_config": {
+                "definition": intent.get("definition", ""),
+                "excludes": [],
+                "examples": intent.get("samples", [])[:3],
+                "covers": [s["name"] for s in sub_intents],
+            },
+            "agent_config": {
+                "type": "chatflow",
+                "nodes": nodes_ac,
+                "footer": f"<b>结束：</b>解决{intent['name']}问题 / 转人工",
+            },
+            "l2_nodes": l2_nodes,
+        })
+
+    edges = []
+    for a in agents:
+        ag_html_id = f"ag-{a['id'].replace('_', '-')}"
+        color = a.get("color", "#6b7280")
+        for node in a["l2_nodes"]:
+            count = node.get("count", 0)
+            if count > 0:
+                edges.append({"from": ag_html_id, "to": node["id"], "count": count, "color": color})
+
+    total_nodes = sum(len(a["l2_nodes"]) for a in agents)
+    n_known = meta.get("n_known", len(results.get("intents", [])))
+    n_new = meta.get("n_new", len(results.get("new_intents", [])))
+
+    rows = []
+    for a in agents:
+        n = len(a["l2_nodes"])
+        badge = f"{n} 节点 ✓" if n <= 2 else (f"{n} 节点 ⚠️" if n <= 3 else f"{n} 节点 ❌")
+        rows.append({
+            "agent_name": a["name"],
+            "color": a["color"],
+            "node_count": n,
+            "badge": badge,
+        })
+
+    l2_total = sum(s["count"] for intent in all_intents for s in intent.get("sub_intents", []))
+    opt_notes = (
+        f"✅ 已知意图 {n_known} 个，新发现意图 {n_new} 个，共 {n_known + n_new} 个 L1 意图\n\n"
+        f"📊 数据基础：{total:,} 条对话 / L2 分类：{l2_total:,} 条"
+    )
+    if n_new > 0:
+        new_names = [i["name"] for i in results.get("new_intents", [])]
+        opt_notes += f"\n\n⚡ 新发现意图：{' · '.join(new_names)}"
+
+    return {
+        "agents": agents,
+        "filter_block": {"types": ["未分类投诉"], "count": 0},
+        "node_summary": {
+            "rows": rows,
+            "total_nodes": total_nodes,
+            "optimization_notes": opt_notes,
+        },
+        "shared_nodes": [],
+        "edges": edges,
+    }
+
+
+def _resolve_results_path(state: dict) -> str:
+    path = state.get("results_path")
+    if not path:
+        raise FileNotFoundError(
+            "No results_path in state. Run classify.py first."
+        )
+    if not os.path.exists(path):
+        raise FileNotFoundError(
+            f"Results file missing: {path}. Run classify.py first."
+        )
+    return path
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Generate poc intent classification HTML report")
+    parser.add_argument("--brand", default="")
+    parser.add_argument("--results", default=None,
+                        help="Override results JSON path (default: read from progress.json)")
+    args = parser.parse_args()
+
+    if args.results:
+        results_path = args.results
+        if not os.path.exists(results_path):
+            raise FileNotFoundError(f"Results file missing: {results_path}")
+    else:
+        state = _read_state()
+        results_path = _resolve_results_path(state)
+
+    with open(results_path) as f:
+        results = json.load(f)
+
+    config = build_config_from_results(results, brand_name=args.brand)
+
+    meta = results.get("meta", {})
+    data_compat = {
+        "_meta": {"total": meta.get("total", 0), "brand": args.brand},
+        "tree": {},
+        "turn_distribution": {},
+        "fuzzy_samples": {},
+    }
+
+    project_id = os.environ.get("SOLVEA_PROJECT_ID", "")
+    html = render_html(config, data_compat, brand=args.brand, project_id=project_id)
+
+    OUTPUTS_DIR.mkdir(parents=True, exist_ok=True)
+    report_path = str(OUTPUTS_DIR / f"report_{timestamp()}.html")
+    with open(report_path, "w", encoding="utf-8") as f:
+        f.write(html)
+    print(f"[gen_report] Report -> {report_path}")
+
+    state = _read_state()
+    state["report_path"] = report_path
+    state["updated_at"] = timestamp()
+    _write_state(state)
+
+    print(f"[gen_report] Done.")
+
+
+if __name__ == "__main__":
+    main()