@topmindspace/tms-skills 0.1.0

package/top-ppt-html/evals/run_evals.py

@@ -0,0 +1,220 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""TopPPT HTML · Evals 运行器（零依赖 · 确定性检查 + 效率度量 + rubric 落地）
+
+用法:
+    python evals/run_evals.py --list                  # 列出 prompt 集（显式/隐式/上下文/负对照）
+    python evals/run_evals.py --budget                # 打印效率预算
+    python evals/run_evals.py --score <报告.html> [--trace <trace.json>] [--json]
+    python evals/run_evals.py --score <报告.html> --rubric <rubric.json>   # 并入定性评分
+
+设计依据（OpenAI《Testing Agent Skills Systematically with Evals》）:
+    一个 prompt → 一次被捕获的运行（trace + artifacts）→ 一小组检查 → 一个可随时间比较的分数。
+    四类目标分别对应：结果（strict 0/0）· 过程（Gate 0 / 轮次）· 风格（rubric）· 效率（轮次 / 工具调用 / 读取字节）。
+
+trace.json 格式（由宿主智能体在跑完一条 prompt 后填写；字段都可选）:
+    {
+      "turns": 3,                     # 交互轮次
+      "toolCalls": 22,                # 工具调用次数（读文件 + 执行脚本）
+      "bytesRead": 41000,             # 智能体读取的总字节数
+      "readFiles": ["references/playbook.md", ...],
+      "gate0ReferenceImage": true,    # 首次交互是否先给出 theme-overview 参考图
+      "artifacts": ["2026-09-15-主题.html"]
+    }
+"""
+from __future__ import annotations
+
+import argparse
+import csv
+import json
+import re
+import subprocess
+import sys
+from pathlib import Path
+
+try:
+    sys.stdout.reconfigure(encoding='utf-8', errors='replace')
+except Exception:
+    pass
+
+ROOT = Path(__file__).resolve().parent.parent
+EVALS = Path(__file__).resolve().parent
+sys.path.insert(0, str(ROOT / 'scripts'))
+import checks_html  # noqa: E402  承载/多样性/组合版式单源
+
+LC = json.loads((ROOT / 'scripts' / 'layout-constants.json').read_text(encoding='utf-8'))
+
+# ── 效率预算（与 SKILL.md 的"预算（硬）"一致；audit_skill.py 校验声明存在） ──────
+BUDGET = {'turns': 3, 'toolCalls': 25, 'bytesRead': 40 * 1024}
+
+# ── 结果/风格检查（确定性部分） ────────────────────────────────────────────────
+CHECK_NAMES = {
+    'artifact': '产物存在且单文件零外链',
+    'strict': 'validate_report --strict 0/0',
+    'model': 'REPORT_MODEL 存在、合法 JSON、mode 一致',
+    'variety': '图表多样性达下限（charts.variety）',
+    'composite': '组合版式比例达下限（research）',
+    'structure': '结构图形/信息图页型被使用（含结构类内容时）',
+    'gate0': '过程：Gate 0 先给参考图',
+    'efficiency': '效率：轮次 / 工具调用 / 读取字节在预算内',
+}
+
+
+def run_validate(html: Path) -> tuple[bool, str]:
+    r = subprocess.run([sys.executable, str(ROOT / 'scripts' / 'validate_report.py'), str(html), '--strict'],
+                       capture_output=True, text=True, encoding='utf-8')
+    out = (r.stdout or '') + (r.stderr or '')
+    tail = [ln for ln in out.splitlines() if ln.startswith('PASS ') or ln.startswith('结论')]
+    return r.returncode == 0, (tail[-1] if tail else out[-200:])
+
+
+def score_artifact(html: Path) -> dict:
+    checks: list[dict] = []
+
+    def add(cid: str, ok: bool, note: str) -> None:
+        checks.append({'id': cid, 'name': CHECK_NAMES[cid], 'ok': ok, 'note': note})
+
+    if not html.exists():
+        add('artifact', False, '文件不存在')
+        return {'artifact': str(html), 'pass': False, 'checks': checks}
+
+    txt = html.read_text(encoding='utf-8')
+    ext = re.findall(r'<(?:link|script|img)[^>]+(?:href|src)=["\']https?://', txt)
+    add('artifact', not ext, f'{len(ext)} 处外链' if ext else f'{len(txt.encode("utf-8")) / 1024:.0f}KB 单文件')
+
+    ok, note = run_validate(html)
+    add('strict', ok, note)
+
+    m = re.search(r'window\.REPORT_MODEL\s*=\s*(\{[\s\S]*?\})\s*;', txt)
+    model = None
+    if m:
+        try:
+            model = json.loads(m.group(1))
+        except json.JSONDecodeError as e:
+            model = None
+            add('model', False, f'JSON 非法: {e}')
+    if m and model is not None:
+        dm = re.search(r'<html[^>]*data-mode="([^"]+)"', txt)
+        consistent = (not model.get('mode')) or (not dm) or model['mode'] == dm.group(1)
+        add('model', consistent, f"mode={model.get('mode')} / data-mode={dm.group(1) if dm else '?'}")
+    elif not m:
+        add('model', False, '未找到 window.REPORT_MODEL')
+
+    mode = (model or {}).get('mode') or 'presentation'
+    per_page = checks_html.chart_types_per_page(txt)
+    used = checks_html.distinct_chart_types(txt)
+    v = checks_html.chart_variety(LC)
+    n_chart_pages = len([1 for ts in per_page if ts])
+    if used:
+        floor = checks_html.variety_floor(mode, n_chart_pages, v)
+        add('variety', len(used) >= floor, f'{len(used)} 种 / 下限 {floor}: {used}')
+    else:
+        add('variety', True, '无图表页（不适用）')
+
+    composite_ratio = checks_html.composite_required(mode, v)
+    if composite_ratio > 0:
+        multi, n_content = checks_html.composite_pages(txt)
+        need = max(1, int(n_content * composite_ratio))
+        add('composite', multi >= need, f'{multi}/{n_content} 页（下限 {need}）')
+
+    # 结构图形/信息图：**只在 architecture 模式硬判**（模式 C 契约 = 图为王，必须有结构主图）；
+    # 其余模式为**信息项**（ok=None，不计入 pass）——信息图页型族对 research/architecture 是
+    # "可用"而非"必须"，纯演示稿/纯数据页没有结构素材，硬判会制造假阴性（属 rubric 的 C3 风格项）。
+    struct_types = {'sankey', 'treemap', 'boxplot', 'network', 'marimekko', 'streamgraph',
+                    'flow', 'tree', 'sequence', 'loop'}
+    has_struct = bool(set(used) & struct_types) or ('class="arch' in txt) or ('class="lane' in txt)
+    if mode == 'architecture':
+        add('structure', has_struct, '架构模式图为王：'
+            + ('已用结构主图' if has_struct else '未用（架构模式必须有分层带/泳道/结构图形）'))
+    else:
+        add('structure', None, '信息图/结构图形：' + ('有' if has_struct else '未使用') + '（本模式为信息项，不计入通过判定）')
+
+    return {'artifact': str(html), 'mode': mode, 'checks': checks}
+
+
+def score_trace(trace_path: Path) -> dict:
+    t = json.loads(trace_path.read_text(encoding='utf-8'))
+    out = []
+    for key, limit in (('turns', BUDGET['turns']), ('toolCalls', BUDGET['toolCalls']),
+                       ('bytesRead', BUDGET['bytesRead'])):
+        v = t.get(key)
+        if v is None:
+            out.append({'id': key, 'ok': None, 'note': '未记录'})
+            continue
+        out.append({'id': key, 'ok': v <= limit, 'note': f'{v} / 上限 {limit}'})
+    g0 = t.get('gate0ReferenceImage')
+    out.append({'id': 'gate0', 'ok': g0 if g0 is not None else None,
+                'note': '首次交互给出参考图' if g0 else ('未给出参考图' if g0 is False else '未记录')})
+    return {'trace': str(trace_path), 'checks': out}
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description='TopPPT HTML Evals 运行器')
+    ap.add_argument('--list', action='store_true', help='列出 prompt 集')
+    ap.add_argument('--budget', action='store_true', help='打印效率预算')
+    ap.add_argument('--score', help='对一份报告产物跑确定性检查')
+    ap.add_argument('--trace', help='过程/效率 trace JSON')
+    ap.add_argument('--rubric', help='已填好的 rubric JSON（定性评分，见 rubric.schema.json）')
+    ap.add_argument('--json', action='store_true', help='机器可读输出')
+    args = ap.parse_args()
+
+    if args.list:
+        rows = list(csv.DictReader((EVALS / 'prompts.csv').read_text(encoding='utf-8').splitlines()))
+        print(f'prompt 集 {len(rows)} 条（显式 / 隐式 / 上下文 / 负对照）\n')
+        for r in rows:
+            flag = '应触发' if r['should_trigger'] == 'true' else '不应触发'
+            print(f"  {r['id']} [{r['kind']:<8}] {flag} · {r['prompt']}")
+            print(f"        期望：{r['expect']}")
+        n_neg = sum(1 for r in rows if r['should_trigger'] != 'true')
+        print(f'\n负对照 {n_neg} 条（捕捉"过于急切地触发技能"的假阳性）')
+        return 0
+
+    if args.budget:
+        print('效率预算（SKILL.md 的"预算（硬）"与 audit_skill.py 同源）:')
+        print(f"  交互轮次 ≤ {BUDGET['turns']} · 工具调用 ≤ {BUDGET['toolCalls']} · "
+              f"读取字节 ≤ {BUDGET['bytesRead'] // 1024}KB")
+        print('  必读文件 = SKILL.md + references/playbook.md（L1）；L2 深度文件按需读、读完即停')
+        return 0
+
+    if not args.score:
+        ap.print_help()
+        return 2
+
+    result = score_artifact(Path(args.score))
+    if args.trace:
+        result['process'] = score_trace(Path(args.trace))
+    if args.rubric:
+        result['rubric'] = json.loads(Path(args.rubric).read_text(encoding='utf-8'))
+
+    def all_ok(checks):
+        return all(c['ok'] for c in checks if c.get('ok') is not None)
+
+    result['pass'] = all_ok(result['checks']) and (
+        all_ok(result['process']['checks']) if 'process' in result else True)
+
+    if args.json:
+        print(json.dumps(result, ensure_ascii=False, indent=2))
+        return 0 if result['pass'] else 1
+
+    print(f"Eval 计分 · {result['artifact']}  (mode={result.get('mode', '?')})")
+    print('-' * 60)
+    for c in result['checks']:
+        mark = '✓' if c['ok'] else ('·' if c['ok'] is None else '✗')
+        print(f"  {mark} [{c['id']}] {c['name']}：{c['note']}")
+    if 'process' in result:
+        print('  ── 过程与效率 ──')
+        for c in result['process']['checks']:
+            mark = '✓' if c['ok'] else ('·' if c['ok'] is None else '✗')
+            print(f"  {mark} [{c['id']}] {c['note']}")
+    if 'rubric' in result:
+        r = result['rubric']
+        print(f"  ── 定性 rubric ── 总分 {r.get('score')} · {'通过' if r.get('overall_pass') else '不通过'}")
+        for c in r.get('checks', []):
+            print(f"    {c.get('score'):>3} [{c.get('id')}] {c.get('notes', '')[:60]}")
+    print('-' * 60)
+    print('结论：' + ('通过' if result['pass'] else '未通过（见上方 ✗ 项）'))
+    return 0 if result['pass'] else 1
+
+
+if __name__ == '__main__':
+    sys.exit(main())

package/top-ppt-html/evals/trace.example.json

@@ -0,0 +1,16 @@
+{
+  "$comment": "trace 示例：跑完一条 prompt 后由宿主智能体填写（字段都可选）。run_evals.py --score <报告.html> --trace evals/trace.example.json 会把过程与效率一并计分。预算单源：SKILL.md「预算（硬）」= 轮次 ≤3 / 工具调用 ≤25 / 读取 ≤40KB。",
+  "promptId": "t02",
+  "turns": 3,
+  "toolCalls": 19,
+  "bytesRead": 34500,
+  "readFiles": [
+    "SKILL.md",
+    "references/playbook.md",
+    "references/components.md",
+    "references/charts.md"
+  ],
+  "gate0ReferenceImage": true,
+  "artifacts": ["2026-09-15-主题.html"],
+  "notes": "轻量路径：1 轮六项问询 + 1 轮交付；只读 L1 + 两份 L2；未整份读取模式模板（用 scaffold_report.py）"
+}

package/top-ppt-html/package.json

@@ -0,0 +1,35 @@
+{
+  "name": "top-ppt-html",
+  "version": "0.1.0",
+  "private": true,
+  "description": "TopPPT HTML：报告 / 演示 / 信息架构图技能。单文件 HTML + 版式保真的可编辑 PPTX；布局骨架 P1–P12；模型单写；strict 0/0 交付。",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/topmindspace/tms-skills.git",
+    "directory": "top-ppt-html"
+  },
+  "type": "commonjs",
+  "engines": {
+    "node": ">=18"
+  },
+  "scripts": {
+    "scaffold": "python scripts/scaffold_report.py",
+    "sync": "python scripts/sync_runtime.py",
+    "audit": "python scripts/audit_styles.py && python scripts/audit_docs.py && python scripts/audit_skill.py && python scripts/audit_css.py",
+    "capture-themes": "node scripts/capture_theme_overview.js",
+    "measure": "python scripts/measure_height.py",
+    "eval": "python evals/run_evals.py",
+    "build-examples": "python scripts/build_examples.py",
+    "regression": "python scripts/regression.py",
+    "negative": "python scripts/negative_tests.py",
+    "package": "python scripts/package_skill.py",
+    "check": "python scripts/package_skill.py --check"
+  },
+  "dependencies": {
+    "pptxgenjs": "^4.0.0"
+  },
+  "devDependencies": {
+    "playwright": "^1.63.0"
+  }
+}