⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
asu-resume-audit-skill
以证据为中心的简历真实性审计与报告生成技能。用户要求核验简历、打假履历、调查候选人经历、分析 GitHub/开源贡献、识别 contributor 到 maintainer/core author 的角色膨胀、核验中外合作办学或学校 title、检查业务指标、梳理公开争议,或者把调查结果制作成 HTML/PDF 时,应主动使用本技能。技能会把简历、截图、链接、帖子、代码仓库和官方记录拆成原子主张,区分事实、矛盾、指控和不可核验信息,并生成不含开盒信息、不把未经证实指控写成事实的自包含证据报告。
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
ASu Resume Skills
| Skill | 用途 | 主要交付 |
|---|---|---|
$asu-resume-skill | 阿酥式履历生成 | 来源映射 JSON、自包含 HTML、A4 PDF、小红书配文 |
$asu-resume-audit-skill | 简历证据审计 | 原子主张、证据矩阵、夸大模式、HTML/PDF 审计报告 |
成品示例
| Skill | JSON 数据 | HTML 成品 | PDF 成品 |
|---|---|---|---|
$asu-resume-skill | resume-data.example.json | asu-resume-example.html | asu-resume-example.pdf |
$asu-resume-audit-skill | audit-report.example.json | asu-resume-audit-example.html | asu-resume-audit-example.pdf |
HTML 均为单文件自包含页面,下载后可直接用浏览器打开;PDF 为同一份 JSON 数据生成的 A4 版本。当前生成示例为 3 页高密度履历,覆盖 3 段经历、7 组核心项目、3 项开源贡献与 7 组技术沉淀;审计示例为 10 页完整报告,包含 16 条原子主张、15 项来源、6 类夸大模式与时间线。
安装
把下面这段直接发给 Codex:
请从这个 GitHub 仓库安装插件,并启用 asu-resume-skill 与 asu-resume-audit-skill:
https://github.com/Claycui828/ASu-resume-skills
安装完成后建议新建一个 Codex 对话,让 Skill 列表重新加载。也可以显式调用:
$asu-resume-skill 读取我的经历,查学校官方 Title,生成 Owner、0→1 和大 Scope 的中文技术履历。
$asu-resume-audit-skill 审计这份简历里的角色强度、Scope 外扩、指标口径和规划时态。
$asu-resume-skill:高密度履历生成
它会完整读取用户提供的简历、飞书文档、截图和项目材料,然后完成:
- 用学校官网、教育主管部门或正式高校联盟核验学校 Title;
- 把真实项目拆成 Data / Model / Training / Serving / Agent / Eval 技术链;
- 为独立负责或主要负责的模块生成边界明确的 Owner、0→1 与架构演进叙事;
- 区分模块 Owner、策略架构 Owner、系统级共建者与规划中方向;
- 输出可截图的中文 HTML、2–3 页 A4 PDF 与小红书配文;
- 用来源标记和指标重算阻止不存在的学历、项目、指标和成果进入成品。
典型用法:
$asu-resume-skill
请读取我的晋升材料和现有简历,保留真实学校、公司、部门和指标。
学校要查官方 Title;项目要有模块级 Owner、0→1、技术链和大 Scope;
最终输出 HTML、PDF 和发布配文。
打招呼示例
原始招聘平台打招呼截图:
$asu-resume-audit-skill:证据优先审计
它不会用“学校不够有名”“项目听起来太大”直接推断造假,而是把每一段拆成可核验的原子主张:
- 学校、学籍、合作项目和学位授予方;
- Contributor、Maintainer、Core Author、Owner、Lead 的角色强度;
- 个人产物、全系统架构、团队结果与个人归因;
- 用户数、转化率、Precision、AUC、延迟与 Benchmark 口径;
- “最年轻、首个、第一”等缺少比较全集的最高级;
- 已交付结果与当前方向、计划、探索之间的时态边界。
每条主张只使用一个状态:已证实、部分证实、存在直接矛盾、缺乏支持 或 暂不可核验。
典型用法:
$asu-resume-audit-skill
请审计这份简历,逐条检查学校 Title、Owner、0→1、核心作者、项目指标和时间线,
生成证据矩阵、自包含 HTML 和 PDF;不要把未经证实的质疑写成事实。
两个 Skill 如何配合
- 用生成 Skill 建立来源映射,输出高密度履历;
- 用审计 Skill 把强表述拆成角色、范围、结果和因果四类主张;
- 对审计报告中的高风险项补设计文档、代码所有权、评审记录、指标看板或学历凭证;
- 将补强后的边界回写简历,再生成最终 HTML/PDF。
真实经历
↓
学校 Title + Owner/0→1 + 技术链 + 大 Scope
↓
原子主张 + 证据等级 + 夸大模式
↓
可发布履历 / 可复核审计报告
事实边界
- 强表达必须限定到真实模块,不能把参与升级成整个项目 Owner;
0→1必须说明“0”指新服务、新训练链路还是新策略节点;- 公司、项目 Star、团队收入和团队指标不会自动成为个人成果;
- 内部指标缺少评测集、分母、时间窗口或看板时,审计结果保持暂不可核验;
- 规划中的 Coding Agent、Memory、Compaction 等方向不能写成已上线收益;
- 审计只描述现有证据,不评价人格,也不把没有公开证据等同于虚假;
- 不要将含姓名、电话、邮箱、内部链接或雇主敏感数据的示例直接提交到公开仓库。
文件结构
ASu-resume-skills/
├── .codex-plugin/
│ └── plugin.json
├── skills/
│ ├── asu-resume-skill/
│ │ ├── SKILL.md
│ │ ├── agents/openai.yaml
│ │ ├── assets/
│ │ ├── references/
│ │ └── scripts/
│ └── asu-resume-audit-skill/
│ ├── SKILL.md
│ ├── agents/openai.yaml
│ ├── assets/
│ ├── references/
│ └── scripts/
├── assets/
│ ├── asu-circle.png
│ └── hr-intro-example.jpg
├── examples/
│ ├── asu-resume-example.html
│ ├── asu-resume-example.pdf
│ ├── asu-resume-audit-example.html
│ ├── asu-resume-audit-example.pdf
│ ├── resume-data.example.json
│ └── audit-report.example.json
├── THIRD_PARTY_LICENSES/
│ └── Hisn00w-ASu-skills-LICENSE
├── LICENSE
└── README.md
开源协议
代码以 MIT License 发布。
// HOW IT'S BUILT
// FILES
13 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
skills/asu-resume-audit-skill/SKILL.md12 KBe6eac0531972upstream ↗
--- name: asu-resume-audit-skill description: 以证据为中心的简历真实性审计与报告生成技能。用户要求核验简历、打假履历、调查候选人经历、分析 GitHub/开源贡献、识别 contributor 到 maintainer/core author 的角色膨胀、核验中外合作办学或学校 title、检查业务指标、梳理公开争议,或者把调查结果制作成 HTML/PDF 时,应主动使用本技能。技能会把简历、截图、链接、帖子、代码仓库和官方记录拆成原子主张,区分事实、矛盾、指控和不可核验信息,并生成不含开盒信息、不把未经证实指控写成事实的自包含证据报告。 --- # ASU 简历真实性审计 Skill 把简历视为一组可以独立检验的主张。目标不是判断一个人“好”或“坏”,而是确定现有证据支持什么、直接反驳什么,以及哪些内容仍然无法核验。 ## 必须交付的结果 除非用户明确缩小范围,否则生成以下内容: 1. `report-data.json`:标准化主张、来源、结论、包装模式、时间线和限制。 2. 自包含 `report.html`:不引用外部样式、脚本、字体或图片。 3. 与同一数据模型一致的 `report.pdf`。 4. 一段简短的对话总结:优先说明最强的支持证据、直接矛盾和未解决缺口。 完成 JSON 后,用 `scripts/render_report.py` 生成 HTML/PDF;交付前用 `scripts/validate_report.py` 校验。 ## 证据与伤害边界 简历审计涉及可识别个人,错误结论可能造成真实的名誉伤害,因此始终遵守以下规则: - 只使用公开且与任务直接相关的信息。不得搜集住址、电话、家庭成员、账号凭证、私人消息或无关个人信息。 - 截图、指控文章、匿名评论和候选人自有主页只能证明“有人这样说过”,不能自动成为独立证据。 - 使用 `公开证据仅支持 contributor,尚未支持 core author` 等精确表述,避免使用 `骗子`、`害虫` 或 `欺诈者` 等人格标签。 - 只有可靠证据与主张直接冲突时才标记 `contradicted`;否则使用 `unsupported`、`partially_corroborated` 或 `unverifiable`。 - 主动提 issue、贡献前端/文档、与维护者建立关系或因此获得正式社区身份,本身不是不当行为。要核验的是简历是否准确描述正式角色、技术深度和个人贡献。 - 不得用学校排名或“双非”身份推断诚信。“双非”是非正式标签,不是造假证据。应核验院校、项目、学籍与学位授予方。 - 性别、外貌、性格、税务或动机推断通常与简历真实性无关;除非权威证据证明其直接关联某条简历主张,否则排除。 - 对当事人有利的重要证据必须与不利证据同等展示。 给结论状态前先阅读 `references/evidence-methodology.md`。核验开源或教育主张前,阅读相应专项参考。 ## 工作流程 ### 1. 确定范围并保存来源集合 列出用户提供的全部材料: - 简历 PDF 或截图; - 文章和社交媒体帖子; - 代码仓库及个人主页链接; - 网页存档; - 用户提供的就业、学校、奖项或付费社群背景。 条件允许时完整阅读链接和文档,以原始分辨率检查图片。页面被安全策略或登录限制阻挡时,说明限制;可使用用户提供的正文或其他公开来源,但不得绕过访问控制。 立即为每个来源分配编号(`S-001`、`S-002`……),并标记来源类型: - `official_record`:官方记录; - `repository_record`:代码仓库记录; - `candidate_self_statement`:候选人自述; - `user_provided_document`:用户提供材料; - `media_or_commentary`:媒体或评论文章; - `anonymous_or_unverified`:匿名或未经核验来源。 ### 2. 把简历拆成原子主张 不要把整段经历作为一条主张。将角色、范围、结果和因果关系分开检验。 例如: > “作为核心作者主导项目从 1.0 到 2.0,性能提升 25%。” 应拆为: 1. 候选人正式或事实上属于核心作者。 2. 候选人领导了 1.0 到 2.0 的迁移。 3. 性能指标确实提升了 25%。 4. 该提升由候选人的工作造成或得到其实质贡献。 每条主张记录: - 简历原文; - 标准化主张; - 主张类别; - 时间范围; - 组织或项目; - 隐含角色与责任范围; - 核验所需证据。 数据格式见 `references/report-schema.md`。 ### 3. 先做确定性的内部一致性检查 联网研究前,先让简历与自身对账: - 重算转化率、增长率和比例; - 比较简历、主页、offer 和帖子中的日期; - 标记重叠的全职或实习时间; - 比较不同平台使用的角色级别; - 找出没有明确比较范围的“最年轻、第一、核心”等最高级; - 区分项目整体指标与个人影响指标; - 检查是否在没有基线或对照组时宣称因果关系。 算术矛盾不依赖外部解释,通常可以给出高置信度。 当材料能够明确给出分子、分母和展示百分比时,在对应的结构化 Claim 中写入: ```json "metric": { "numerator": 200, "denominator": 2000, "displayed_percent": 10 } ``` 交由确定性校验器重算。不要从只有“转化率 38%”的自然语言中猜测分母。数字关系不一致只说明当前口径下的数值不一致,不自动等于候选人造假;若可能存在未披露的分母,应在分析中说明这一限制。 当材料明确给出基线、结果、变化类型和展示变化值时,也在同一个 `metric` 对象中记录 `baseline`、`result`、`change_type` 与 `displayed_change`。`relative_percent` 按 `(result - baseline) / baseline * 100` 重算,例如 100 到 125 是 `25%`;`percentage_points` 直接相减,例如 60% 到 70% 是 `10` 个百分点,而不是 `10%` 的相对变化(相对变化约为 16.67%)。不要从自然语言猜测变化类型或基线。相对变化的 baseline 为零时应说明无法计算;数字关系不一致同样只说明已提供数值之间的关系,不自动等于候选人造假。 ### 4. 用最强来源核验每条主张 技术与机构主张优先使用一手资料。 #### 开源与 GitHub 主张 先阅读 `references/open-source-audit.md`,再检查: - 作者 PR 和 issue 搜索; - merged、closed-unmerged、open、duplicate、superseded 的区别; - 修改文件、代码所有权、评审记录和 release note; - README 致谢、maintainer 列表、CODEOWNERS、组织成员与正式任命; - 贡献类型:核心引擎、功能、缺陷修复、前端、网站、文档、测试、生成代码、评论或仅 issue; - 贡献时间与所声称版本/架构阶段的关系; - star、排名、采用率和项目成功是否早于候选人的贡献。 GitHub contribution 数量不等于合并到生产的代码;某个子项目的正式 committer 身份也不等于整个核心引擎的作者身份。 #### 就业、offer 与内部项目主张 优先使用雇主记录、offer 元数据、主管确认、内部设计文档、代码所有权、上线记录或指标看板。无法取得时标记 `unverifiable`,不能直接判假。 遇到 `owner`、`0-to-1`、`core author`、`architect` 或 `lead` 时追问: - 谁授予或分配了这个角色? - 实际负责哪个子系统? - 哪些决策由候选人本人提出? - 哪些产物通过评审并进入生产? - 哪个结果可以归因于个人而不是团队? #### 学校与中外合作项目主张 阅读 `references/education-branding-audit.md`,分别核验: - 录取和注册学籍所在院校; - 实际就读校区或项目; - 海外合作院校; - 交换或访问身份; - 中外合作项目状态; - 学位授予院校; - 最终取得的学位、文凭或证书。 只有在官方项目和学历证明允许的范围内,才能使用合作院校品牌。不得因为学校不够知名就推断造假。 #### 产品、创业与付费用户指标 要求定义注册用户、活跃用户、付费用户、转化事件、分母、时间窗口、收入、退款和测试账号。重算所有可见比例。第三方流量估算不能证明内部支付数据或项目归属。 #### 奖项与“最年轻/首个/第一”主张 查找官方获奖名单和明确的比较集合。没有公开排名或可穷举比较范围时,即使底层奖项或身份真实,最高级仍应标记 `unsupported`。 ### 5. 检查简历包装与夸大模式 阅读 `references/inflation-patterns.md`。必须先完成逐条证据核验,再判断是否存在模式。常见模式包括: - contributor 升级为 maintainer/core author; - 网站或文档贡献被表述为核心引擎深度; - 项目 star、品牌和排名转移到个人; - 社交或可见度工作被用来暗示技术权威; - 复述完整系统架构代替个人产物证明; - 团队结果和采用率归因于个人; - 分母切换和伪精确指标; - 在多个自有平台重复自授 title; - 将合作学校或海外项目呈现为学位主体; - 时间压缩及重复的 `owner/0-to-1` 表述。 发现模式只是调查信号,不能单独作为造假证据。 ### 6. 分配证据状态 每条主张只使用一个状态: - `corroborated`:强独立证据支持主张的关键内容。 - `partially_corroborated`:真实事实存在,但角色、范围、因果或幅度超过证据。 - `contradicted`:可靠证据与主张直接冲突。 - `unsupported`:合理搜索后仍没有足够证据。 - `unverifiable`:只有当前不可取得的内部或私人记录才能核验。 同时记录置信度(`high`、`medium`、`low`),并说明什么证据可以改变结论。 ### 7. 建立报告数据 按 `references/report-schema.md` 创建 `report-data.json`。引用保持简短,每个重要结论都要指向来源。指控必须带归属,例如: > `文章作者指控……;当前材料缺少原始截图,尚未独立验证。` 不要直接写: > `当事人伪造了……` 除非直接、权威的证据已经证明该结论。 ### 8. 生成 HTML 与 PDF 在技能目录运行: ```bash python3 scripts/render_report.py \ --input /absolute/path/report-data.json \ --html /absolute/path/report.html \ --pdf /absolute/path/report.pdf ``` HTML 必须保持自包含并使用附带的纸张式证据报告设计,至少包含: 1. 范围与限制; 2. 证据状态卡片; 3. 调查流程图; 4. 最高风险主张; 5. 完整主张-证据矩阵; 6. 包装模式及反向证据; 7. 时间线; 8. 来源和下一步核验。 ### 9. 校验并视觉检查 运行: ```bash python3 scripts/validate_report.py \ --data /absolute/path/report-data.json \ --html /absolute/path/report.html \ --pdf /absolute/path/report.pdf ``` 如果安装了 `pdftoppm`,将 PDF 渲染为 PNG 并检查每一页,确认没有文字裁切、中文缺字、表格破裂或元素重叠。发现问题后先修改再交付。 ### 10. 使用校准后的语言交付 结论顺序为: - 最强的已证实事实; - 最强的直接矛盾; - 最大的不可核验缺口。 不要复述文章中的煽动性措辞。明确区分 `字面虚构`、`角色/范围膨胀`、`缺乏支持的营销表述` 和 `仅仅缺少证据`。 ## 附带参考资料 - `references/evidence-methodology.md`:来源等级与状态判断规则。 - `references/open-source-audit.md`:GitHub/Apache 贡献和角色审计。 - `references/education-branding-audit.md`:院校、中外合作、交换和学位表述。 - `references/inflation-patterns.md`:防御性简历包装分类。 - `references/report-schema.md`:报告生成器读取的 JSON 格式。 - `references/asu-case-study.md`:基于公开记录和用户材料的有限案例;不得将其视为永恒或全面事实。 ## 附带工具 - `scripts/render_report.py`:确定性 HTML/PDF 生成器。 - `scripts/validate_report.py`:数据结构、证据引用、HTML 和 PDF 冒烟检查。 - `assets/report_template.html`:自包含中文 HTML 报告模板。 - `evals/evals.json`:中文测试任务和预期行为。skills/asu-resume-audit-skill/scripts/render_report.py18 KB054a299612faupstream ↗
#!/usr/bin/env python3 """从统一 JSON 数据生成中文 HTML 和 PDF 简历证据审计报告。""" from __future__ import annotations import argparse import json import re from collections import Counter from pathlib import Path from typing import Any from xml.sax.saxutils import escape STATUS_ZH = { "corroborated": "已证实", "partially_corroborated": "部分证实", "contradicted": "存在直接矛盾", "unsupported": "缺乏支持", "unverifiable": "暂不可核验", } RISK_ZH = {"low": "低", "medium": "中", "high": "高", "critical": "严重"} CONFIDENCE_ZH = {"low": "低", "medium": "中", "high": "高"} def normalize_pdf_text(value: Any) -> str: """把 PDF 中容易造成断行或字体问题的破折号统一为 ASCII。""" text = "" if value is None else str(value) return ( text.replace("‐", "-") .replace("‑", "-") .replace("‒", "-") .replace("–", "-") .replace("—", "-") .replace("−", "-") .replace("·", " / ") ) def ptext(value: Any) -> str: return escape(normalize_pdf_text(value)).replace("\n", "<br/>") def load_data(path: Path) -> dict[str, Any]: with path.open("r", encoding="utf-8") as handle: return json.load(handle) def render_html(data: dict[str, Any], template_path: Path, output_path: Path) -> None: template = template_path.read_text(encoding="utf-8") payload = json.dumps(data, ensure_ascii=False, separators=(",", ":")) payload = payload.replace("</", "<\\/") if "__REPORT_JSON__" not in template: raise ValueError(f"HTML 模板缺少 __REPORT_JSON__ 占位符:{template_path}") html = template.replace("__REPORT_JSON__", payload) output_path.parent.mkdir(parents=True, exist_ok=True) output_path.write_text(html, encoding="utf-8") def build_pdf(data: dict[str, Any], output_path: Path) -> None: try: from reportlab.lib import colors from reportlab.lib.enums import TA_CENTER, TA_LEFT from reportlab.lib.pagesizes import A4 from reportlab.lib.styles import ParagraphStyle, getSampleStyleSheet from reportlab.lib.units import mm from reportlab.pdfbase import pdfmetrics from reportlab.pdfbase.cidfonts import UnicodeCIDFont from reportlab.platypus import ( LongTable, PageBreak, Paragraph, SimpleDocTemplate, Spacer, Table, TableStyle, ) except ImportError as exc: raise SystemExit("缺少 reportlab。请运行:python3 -m pip install reportlab") from exc output_path.parent.mkdir(parents=True, exist_ok=True) try: pdfmetrics.registerFont(UnicodeCIDFont("STSong-Light")) font_name = "STSong-Light" except Exception: font_name = "Helvetica" palette = { "ivory": colors.HexColor("#FAF9F5"), "slate": colors.HexColor("#141413"), "clay": colors.HexColor("#D97757"), "oat": colors.HexColor("#E3DACC"), "olive": colors.HexColor("#788C5D"), "gray100": colors.HexColor("#F0EEE6"), "gray300": colors.HexColor("#D1CFC5"), "gray500": colors.HexColor("#87867F"), "gray700": colors.HexColor("#3D3D3A"), "rust": colors.HexColor("#B04A3F"), "info": colors.HexColor("#5C7CA3"), "warning": colors.HexColor("#C78E3F"), } styles = getSampleStyleSheet() body = ParagraphStyle( "中文正文", parent=styles["BodyText"], fontName=font_name, fontSize=9.4, leading=14, textColor=palette["gray700"], wordWrap="CJK", spaceAfter=5, ) small = ParagraphStyle( "中文小字", parent=body, fontSize=7.8, leading=11, textColor=palette["gray500"], ) title = ParagraphStyle( "中文标题", parent=body, fontSize=24, leading=31, textColor=palette["slate"], alignment=TA_LEFT, spaceAfter=8, ) subtitle = ParagraphStyle( "中文副标题", parent=body, fontSize=12, leading=18, textColor=palette["gray700"], spaceAfter=12, ) h2 = ParagraphStyle( "中文二级标题", parent=body, fontSize=16, leading=22, textColor=palette["slate"], spaceBefore=12, spaceAfter=9, keepWithNext=True, ) h3 = ParagraphStyle( "中文三级标题", parent=body, fontSize=12, leading=17, textColor=palette["slate"], spaceBefore=7, spaceAfter=5, keepWithNext=True, ) badge = ParagraphStyle( "标签", parent=small, alignment=TA_CENTER, textColor=palette["slate"], ) cell = ParagraphStyle("表格正文", parent=body, fontSize=7.8, leading=11, spaceAfter=0) cell_small = ParagraphStyle("表格小字", parent=small, fontSize=7.0, leading=9.5, spaceAfter=0) metadata = data.get("metadata", {}) summary = data.get("summary", {}) claims = data.get("claims", []) sources = data.get("sources", []) patterns = data.get("patterns", []) timeline = data.get("timeline", []) next_steps = data.get("next_steps", []) counts = Counter(c.get("status") for c in claims) doc = SimpleDocTemplate( str(output_path), pagesize=A4, rightMargin=17 * mm, leftMargin=17 * mm, topMargin=17 * mm, bottomMargin=17 * mm, title=normalize_pdf_text(metadata.get("title", "简历真实性审计报告")), author=normalize_pdf_text(metadata.get("analyst", "asu-resume-audit-skill")), ) story: list[Any] = [] story.append(Paragraph("证据优先 / 简历主张审计", small)) story.append(Paragraph(ptext(metadata.get("title", "简历真实性审计报告")), title)) story.append(Paragraph(ptext(metadata.get("subject", "未命名审计对象")), subtitle)) meta_line = " · ".join( [ "范围:" + normalize_pdf_text(metadata.get("scope", "未说明")), "生成时间:" + normalize_pdf_text(metadata.get("generated_at", "未说明")), "分析者:" + normalize_pdf_text(metadata.get("analyst", "未说明")), ] ) story.append(Paragraph(ptext(meta_line), small)) story.append(Spacer(1, 7 * mm)) story.append(Paragraph("证据结论总览", h2)) status_order = [ "corroborated", "partially_corroborated", "contradicted", "unsupported", "unverifiable", ] metric_cells = [] metric_colors = [palette["olive"], palette["info"], palette["rust"], palette["warning"], palette["gray500"]] for status, color in zip(status_order, metric_colors): metric_cells.append( Table( [[Paragraph(str(counts.get(status, 0)), ParagraphStyle("数字", parent=title, fontSize=20, leading=22, alignment=TA_CENTER))], [Paragraph(ptext(STATUS_ZH[status]), badge)]], colWidths=[30 * mm], rowHeights=[13 * mm, 8 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-1, -1), colors.white), ("BOX", (0, 0), (-1, -1), 0.7, palette["gray300"]), ("LINEBEFORE", (0, 0), (0, -1), 3, color), ("VALIGN", (0, 0), (-1, -1), "MIDDLE"), ("LEFTPADDING", (0, 0), (-1, -1), 4), ("RIGHTPADDING", (0, 0), (-1, -1), 4), ]), ) ) story.append(Table([metric_cells], colWidths=[33 * mm] * 5, style=TableStyle([("VALIGN", (0, 0), (-1, -1), "TOP")]))) story.append(Spacer(1, 4 * mm)) conclusion_box = Table( [[Paragraph(ptext(summary.get("conclusion", "尚未形成结论。")), body)]], colWidths=[168 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-1, -1), colors.white), ("BOX", (0, 0), (-1, -1), 0.7, palette["gray300"]), ("LINEBEFORE", (0, 0), (0, -1), 3, palette["clay"]), ("LEFTPADDING", (0, 0), (-1, -1), 10), ("RIGHTPADDING", (0, 0), (-1, -1), 10), ("TOPPADDING", (0, 0), (-1, -1), 8), ("BOTTOMPADDING", (0, 0), (-1, -1), 8), ]), ) story.append(conclusion_box) if metadata.get("disclaimer"): story.append(Paragraph(ptext(metadata["disclaimer"]), small)) if summary.get("limitations"): story.append(Paragraph("已知限制", h3)) for item in summary["limitations"]: story.append(Paragraph("- " + ptext(item), body)) story.append(Paragraph("审计流程", h2)) flow_data = [[ Paragraph("解析简历<br/><font size='7'>文本 / 图片 / PDF</font>", badge), Paragraph("拆分主张<br/><font size='7'>角色 / 范围 / 指标</font>", badge), Paragraph("检索证据<br/><font size='7'>官方 / GitHub / 档案</font>", badge), Paragraph("交叉核验<br/><font size='7'>支持 / 矛盾 / 缺口</font>", badge), Paragraph("生成报告<br/><font size='7'>JSON / HTML / PDF</font>", badge), ]] story.append(Table(flow_data, colWidths=[33.5 * mm] * 5, rowHeights=[17 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-2, -1), colors.white), ("BACKGROUND", (-1, 0), (-1, -1), palette["oat"]), ("BOX", (0, 0), (-1, -1), 0.7, palette["gray300"]), ("INNERGRID", (0, 0), (-1, -1), 0.7, palette["gray300"]), ("VALIGN", (0, 0), (-1, -1), "MIDDLE"), ]))) high_ids = set(summary.get("highest_risk", [])) high_claims = [c for c in claims if c.get("id") in high_ids or c.get("risk") == "critical"] if not high_claims: high_claims = claims[:5] story.append(Paragraph("最高风险主张", h2)) high_rows = [[ Paragraph("编号", cell_small), Paragraph("简历主张", cell_small), Paragraph("结论", cell_small), Paragraph("风险", cell_small), Paragraph("核心分析", cell_small), ]] for claim in high_claims: high_rows.append([ Paragraph(ptext(claim.get("id")), cell_small), Paragraph(ptext(claim.get("resume_text") or claim.get("normalized_claim")), cell), Paragraph(ptext(STATUS_ZH.get(claim.get("status"), claim.get("status"))), cell_small), Paragraph(ptext(RISK_ZH.get(claim.get("risk"), claim.get("risk"))), cell_small), Paragraph(ptext(claim.get("analysis")), cell), ]) story.append(LongTable(high_rows, repeatRows=1, colWidths=[13 * mm, 42 * mm, 21 * mm, 13 * mm, 79 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-1, 0), palette["gray100"]), ("BOX", (0, 0), (-1, -1), 0.6, palette["gray300"]), ("INNERGRID", (0, 0), (-1, -1), 0.35, palette["gray300"]), ("VALIGN", (0, 0), (-1, -1), "TOP"), ("LEFTPADDING", (0, 0), (-1, -1), 5), ("RIGHTPADDING", (0, 0), (-1, -1), 5), ("TOPPADDING", (0, 0), (-1, -1), 5), ("BOTTOMPADDING", (0, 0), (-1, -1), 5), ]))) story.append(PageBreak()) story.append(Paragraph("完整主张与证据矩阵", h2)) for claim in claims: if claim.get("page_break_before"): story.append(PageBreak()) evidence_for = "、".join(claim.get("evidence_for", [])) or "无" evidence_against = "、".join(claim.get("evidence_against", [])) or "无" steps = claim.get("verification_steps", []) block = [ [Paragraph(ptext(claim.get("id")), cell_small), Paragraph(ptext(claim.get("normalized_claim")), cell)], [Paragraph("状态", cell_small), Paragraph(ptext(STATUS_ZH.get(claim.get("status"), claim.get("status"))), cell)], [Paragraph("风险 / 置信度", cell_small), Paragraph(ptext(f"{RISK_ZH.get(claim.get('risk'), claim.get('risk'))} / {CONFIDENCE_ZH.get(claim.get('confidence'), claim.get('confidence'))}"), cell)], [Paragraph("证据", cell_small), Paragraph(ptext(f"支持:{evidence_for};反向:{evidence_against}"), cell)], [Paragraph("分析", cell_small), Paragraph(ptext(claim.get("analysis")), cell)], [Paragraph("下一步", cell_small), Paragraph("<br/>".join("- " + ptext(x) for x in steps) or "无", cell)], ] claim_table = Table(block, colWidths=[27 * mm, 141 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (0, -1), palette["gray100"]), ("BOX", (0, 0), (-1, -1), 0.6, palette["gray300"]), ("INNERGRID", (0, 0), (-1, -1), 0.35, palette["gray300"]), ("VALIGN", (0, 0), (-1, -1), "TOP"), ("LEFTPADDING", (0, 0), (-1, -1), 6), ("RIGHTPADDING", (0, 0), (-1, -1), 6), ("TOPPADDING", (0, 0), (-1, -1), 5), ("BOTTOMPADDING", (0, 0), (-1, -1), 5), ])) story.extend([claim_table, Spacer(1, 3 * mm)]) story.append(Paragraph("识别到的包装与夸大模式", h2)) for pattern in patterns: if pattern.get("page_break_before"): story.append(PageBreak()) signals = ";".join(pattern.get("signals", [])) or "无" pattern_table = Table([ [Paragraph(ptext(pattern.get("name")), h3)], [Paragraph(ptext(pattern.get("description")), body)], [Paragraph(ptext("观察信号:" + signals), body)], [Paragraph(ptext("反向证据/合理解释:" + (pattern.get("counter_evidence") or "无")), body)], [Paragraph(ptext("置信度:" + CONFIDENCE_ZH.get(pattern.get("confidence"), pattern.get("confidence"))), small)], ], colWidths=[168 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-1, -1), colors.white), ("BOX", (0, 0), (-1, -1), 0.6, palette["gray300"]), ("LINEBEFORE", (0, 0), (0, -1), 3, palette["clay"]), ("LEFTPADDING", (0, 0), (-1, -1), 8), ("RIGHTPADDING", (0, 0), (-1, -1), 8), ("TOPPADDING", (0, 0), (-1, -1), 4), ("BOTTOMPADDING", (0, 0), (-1, -1), 4), ])) story.extend([pattern_table, Spacer(1, 3 * mm)]) if timeline: story.append(Paragraph("时间线", h2)) time_rows = [[Paragraph("日期", cell_small), Paragraph("事件", cell_small), Paragraph("详情", cell_small), Paragraph("来源", cell_small)]] for item in timeline: time_rows.append([ Paragraph(ptext(item.get("date")), cell_small), Paragraph(ptext(item.get("title")), cell), Paragraph(ptext(item.get("detail")), cell), Paragraph(ptext("、".join(item.get("source_ids", [])) or "无"), cell_small), ]) story.append(LongTable(time_rows, repeatRows=1, colWidths=[28 * mm, 39 * mm, 78 * mm, 23 * mm], style=TableStyle([ ("BACKGROUND", (0, 0), (-1, 0), palette["gray100"]), ("BOX", (0, 0), (-1, -1), 0.6, palette["gray300"]), ("INNERGRID", (0, 0), (-1, -1), 0.35, palette["gray300"]), ("VALIGN", (0, 0), (-1, -1), "TOP"), ("LEFTPADDING", (0, 0), (-1, -1), 5), ("RIGHTPADDING", (0, 0), (-1, -1), 5), ("TOPPADDING", (0, 0), (-1, -1), 5), ("BOTTOMPADDING", (0, 0), (-1, -1), 5), ]))) story.append(Paragraph("下一步核验", h2)) for index, item in enumerate(next_steps, 1): story.append(Paragraph(ptext(f"{index}. {item}"), body)) story.append(Paragraph("来源与证据等级", h2)) for source in sources: label = f"{source.get('id')} · {source.get('title')}" url = normalize_pdf_text(source.get("url", "")) if url and re.match(r"^https?://", url): heading = f'<link href="{escape(url)}" color="#D97757">{ptext(label)}</link>' else: heading = ptext(label) source_block = [ Paragraph(heading, h3), Paragraph(ptext(" · ".join(str(x) for x in [source.get("type"), source.get("publisher"), source.get("accessed_at"), "等级 " + str(source.get("reliability", "?"))] if x)), small), ] if source.get("notes"): source_block.append(Paragraph(ptext(source.get("notes")), body)) story.extend(source_block) def decorate_page(canvas, document): canvas.saveState() canvas.setStrokeColor(palette["gray300"]) canvas.setLineWidth(0.5) canvas.line(17 * mm, 12 * mm, 193 * mm, 12 * mm) canvas.setFont(font_name, 7) canvas.setFillColor(palette["gray500"]) canvas.drawString(17 * mm, 8 * mm, "asu-resume-audit-skill / 证据快照") canvas.drawRightString(193 * mm, 8 * mm, f"第 {document.page} 页") canvas.restoreState() doc.build(story, onFirstPage=decorate_page, onLaterPages=decorate_page) def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description="生成中文简历真实性审计 HTML/PDF") parser.add_argument("--input", required=True, type=Path, help="报告 JSON 数据") parser.add_argument("--html", required=True, type=Path, help="HTML 输出路径") parser.add_argument("--pdf", required=True, type=Path, help="PDF 输出路径") parser.add_argument("--template", type=Path, help="可选 HTML 模板路径") return parser.parse_args() def main() -> None: args = parse_args() skill_root = Path(__file__).resolve().parents[1] template = args.template or skill_root / "assets" / "report_template.html" data = load_data(args.input) render_html(data, template, args.html) build_pdf(data, args.pdf) print(f"HTML 已生成:{args.html.resolve()}") print(f"PDF 已生成:{args.pdf.resolve()}") if __name__ == "__main__": main()skills/asu-resume-audit-skill/scripts/validate_report.py7.3 KB1794276991aeupstream ↗
#!/usr/bin/env python3 """校验报告 JSON、HTML 自包含性与 PDF 基本完整性。""" from __future__ import annotations import argparse import json import math import re import sys from pathlib import Path from urllib.parse import urlparse ALLOWED_STATUSES = { "corroborated", "partially_corroborated", "contradicted", "unsupported", "unverifiable", } ALLOWED_RISKS = {"low", "medium", "high", "critical"} ALLOWED_CONFIDENCE = {"low", "medium", "high"} def fail(errors: list[str], message: str) -> None: errors.append(message) def validate_metric(claim: dict, errors: list[str]) -> None: metric = claim.get("metric") if metric is None: return cid = claim.get("id", "<无编号>") if not isinstance(metric, dict): fail(errors, f"{cid} metric 必须是对象") return ratio_keys = {"numerator", "denominator", "displayed_percent"} if ratio_keys <= set(metric): try: expected = float(metric["numerator"]) / float(metric["denominator"]) * 100 displayed = float(metric["displayed_percent"]) except (TypeError, ValueError, ZeroDivisionError): fail(errors, f"{cid} metric 指标字段无法计算") else: if not math.isclose(expected, displayed, abs_tol=0.5): fail(errors, f"{cid} metric 百分比与分子分母不一致") change_keys = {"baseline", "result", "change_type", "displayed_change"} if not change_keys <= set(metric): return try: baseline = float(metric["baseline"]) result = float(metric["result"]) displayed_change = float(metric["displayed_change"]) except (TypeError, ValueError): fail(errors, f"{cid} metric 变化字段无法计算") return change_type = metric["change_type"] if change_type == "relative_percent": if baseline == 0: fail(errors, f"{cid} metric 相对百分比变化的 baseline 不能为 0") return expected_change = (result - baseline) / baseline * 100 elif change_type == "percentage_points": expected_change = result - baseline else: fail(errors, f"{cid} metric change_type 非法:{change_type}") return if not math.isclose(expected_change, displayed_change, abs_tol=0.5): fail(errors, f"{cid} metric 变化值与基线结果不一致") def validate_data(data: dict, errors: list[str]) -> None: metadata = data.get("metadata", {}) for key in ("title", "subject", "generated_at"): if not metadata.get(key): fail(errors, f"metadata.{key} 缺失") sources = data.get("sources") claims = data.get("claims") if not isinstance(sources, list) or not sources: fail(errors, "sources 必须是非空数组") sources = [] if not isinstance(claims, list) or not claims: fail(errors, "claims 必须是非空数组") claims = [] source_ids = {s.get("id") for s in sources if s.get("id")} if len(source_ids) != len(sources): fail(errors, "source id 缺失或重复") for source in sources: sid = source.get("id", "<无编号>") url = source.get("url") if url: parsed = urlparse(str(url)) if parsed.scheme not in {"http", "https"} or not parsed.netloc: fail(errors, f"{sid} URL 必须使用 http 或 https:{url}") claim_ids = {c.get("id") for c in claims if c.get("id")} if len(claim_ids) != len(claims): fail(errors, "claim id 缺失或重复") for claim in claims: cid = claim.get("id", "<无编号>") for key in ("normalized_claim", "status", "risk", "confidence", "analysis"): if not claim.get(key): fail(errors, f"{cid} 缺少 {key}") if claim.get("status") not in ALLOWED_STATUSES: fail(errors, f"{cid} status 非法:{claim.get('status')}") if claim.get("risk") not in ALLOWED_RISKS: fail(errors, f"{cid} risk 非法:{claim.get('risk')}") if claim.get("confidence") not in ALLOWED_CONFIDENCE: fail(errors, f"{cid} confidence 非法:{claim.get('confidence')}") validate_metric(claim, errors) for sid in claim.get("evidence_for", []) + claim.get("evidence_against", []): if sid not in source_ids: fail(errors, f"{cid} 引用了不存在的 source id:{sid}") for pattern in data.get("patterns", []): for cid in pattern.get("claim_ids", []): if cid not in claim_ids: fail(errors, f"模式 {pattern.get('name')} 引用了不存在的 claim id:{cid}") for item in data.get("timeline", []): for sid in item.get("source_ids", []): if sid not in source_ids: fail(errors, f"时间线 {item.get('title')} 引用了不存在的 source id:{sid}") def validate_html(path: Path, data: dict, errors: list[str]) -> None: if not path.exists() or path.stat().st_size < 10_000: fail(errors, f"HTML 不存在或体积异常:{path}") return html = path.read_text(encoding="utf-8") if "__REPORT_JSON__" in html: fail(errors, "HTML 仍包含未替换的数据占位符") if re.search(r"<script[^>]+src\s*=", html, re.I): fail(errors, "HTML 引用了外部脚本,不是自包含文件") if re.search(r"<link[^>]+rel=[\"']?stylesheet", html, re.I): fail(errors, "HTML 引用了外部样式表,不是自包含文件") subject = str(data.get("metadata", {}).get("subject", "")) if subject and subject not in html: fail(errors, "HTML 中未找到报告对象名称") for claim in data.get("claims", [])[:3]: if str(claim.get("id")) not in html: fail(errors, f"HTML 中未找到 claim id:{claim.get('id')}") def validate_pdf(path: Path, errors: list[str]) -> None: if not path.exists() or path.stat().st_size < 5_000: fail(errors, f"PDF 不存在或体积异常:{path}") return with path.open("rb") as handle: if handle.read(4) != b"%PDF": fail(errors, "PDF 文件头无效") try: from pypdf import PdfReader reader = PdfReader(str(path)) if not reader.pages: fail(errors, "PDF 没有页面") except ImportError: pass except Exception as exc: fail(errors, f"PDF 解析失败:{exc}") def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description="校验中文简历真实性审计报告") parser.add_argument("--data", required=True, type=Path) parser.add_argument("--html", required=True, type=Path) parser.add_argument("--pdf", required=True, type=Path) return parser.parse_args() def main() -> None: args = parse_args() errors: list[str] = [] try: data = json.loads(args.data.read_text(encoding="utf-8")) except Exception as exc: print(f"JSON 读取失败:{exc}", file=sys.stderr) raise SystemExit(1) validate_data(data, errors) validate_html(args.html, data, errors) validate_pdf(args.pdf, errors) if errors: print("校验失败:", file=sys.stderr) for error in errors: print(f"- {error}", file=sys.stderr) raise SystemExit(1) print("校验通过:JSON 结构、证据引用、HTML 自包含性和 PDF 基本完整性均正常。") if __name__ == "__main__": main()skills/asu-resume-audit-skill/references/asu-case-study.md5.2 KB4ab1d9c4bf40upstream ↗
# ASU 公开证据案例研究 本案例演示如何审计争议简历,而不把指控文章直接当成事实。它是有范围限制的工作样例,不是对任何人的永久结论。 ## 来源边界 创建技能时可用材料包括: - 用户提供的两张简历截图; - 用户粘贴的一篇公开公众号质疑文章全文,但多数内嵌证据图片缺失; - 2026-08-13 可访问的公开 GitHub 项目、个人主页、PR 与 Apache 名册记录。 目标公众号页面因站点安全限制无法通过浏览器打开;用户随后提供了完整正文。依赖缺失图片的主张没有得到独立核验。 ## 截图中可见的主要简历主张 - 当事人是 DeerFlow `项目 owner 核心作者之一`,并主导 1.0 到 2.0 的架构升级; - 当事人是 Apache Committer,且 `18 岁成为全球最年轻 Apache Committer`; - 多段短期实习中反复出现 `项目 owner`、`完全负责 0 到 1` 等表述; - Thumbnail AI 有 2,000+ 注册用户、200+ 付费用户,付费转化率约 38%; - Krene-Art 有 1,200+ 用户、150+ 付费用户,付费转化率约 42%; - 多个内部 AI4SE/AgentScaling/Tracer 项目声称取得 Benchmark 与成本改进。 截图未展示完整教育条目,因此无法仅凭图片判定中外合作项目或海外院校 Title 的表述。 ## 公开核验结果 ### DeerFlow:参与事实与核心作者主张 公开记录支持确有参与: - 审计时 GitHub 作者搜索返回 5 个 `bytedance/deer-flow` PR; - PR #1077 已合并,内容为有边界的前端导航/状态修复; - PR #1081 与 #1064 涉及中间件、后端或内存管理尝试,但关闭未合并。 相关链接: - https://github.com/bytedance/deer-flow/pulls?q=is%3Apr+author%3ALofiSu - https://github.com/bytedance/deer-flow/pull/1077 - https://github.com/bytedance/deer-flow/pull/1081 - https://github.com/bytedance/deer-flow/pull/1064 公开 README 将 2.0 描述为从头重写,并在致谢中列出 Daniel Walnut 与 Henry Li 为核心作者;该段未列出 LofiSu: - https://github.com/bytedance/deer-flow#acknowledgments 校准结论:Contributor 活动已证实;“核心作者并主导 1.0 到 2.0 重写”的更强表述与公开记录存在张力,未得到公开证实。仍可能存在内部工作,因此状态应为 `partially_corroborated`,而不是断言完全没有核心工作。 ### Apache 角色与技术范围 Apache 公开名册确认 `LofiSu` 是 Fory Committer: - https://people.apache.org/phonebook.html?unix=fory 公开 GitHub 搜索显示有实质 Apache 活动,但多数集中在 `apache/fory-site` 官网、前端与文档;可见的主引擎工作很有限。这不推翻正式 Committer 头衔,只限制公开证据能够支持的“核心序列化引擎深度”。 `全球最年轻` 缺乏 ASF 全局年龄排名。审计时 Apache 记录显示 2025 年 4 月新增 Committer,而个人 GitHub 页面写明出生于 2006 年 1 月,这使“18 岁”出现时间线疑问;正式邀请日期可能早于名册更新。负责任的状态是 `unsupported` 或 `partially_corroborated`,而非在任命日期确定前直接写 `contradicted`。 ### 确定性的指标矛盾 - 200 / 2,000 = 10%,不是 38%; - 150 / 1,200 = 12.5%,不是 42%。 百分比可能使用更窄漏斗分母,但简历没有披露。可见呈现内部不一致,存在误导风险;这是高置信度结论。 ### 内部 AI、Offer 与就业主张 AI Infra 所有权、内部 Benchmark、成本下降、榜单、Offer 类型、薪酬和具体数据工作需要雇主或内部记录。质疑文章提出了相关指控,但粘贴文本没有包含支撑截图,因此样例报告必须标记为 `unverifiable`。 ### 学历品牌 用户指控国内院校与海外合作项目 Title 被视觉绑定。现有截图没有展示教育条目,也没有官方项目或学位记录,不能把该指控重复成事实。取得完整条目后使用 `education-branding-audit.md`。 ### 主动贡献与社交策略 公开记录显示当事人积极参与官网/前端工作并与 Maintainer 互动。这些行为可以正当地带来信任与正式角色。审计关注的不是“靠社交拿 Title”,而是后续简历是否: - 省略官网/前端范围; - 用项目级正式头衔暗示核心引擎作者身份; - 用知名关系或项目品牌替代个人工件。 ## 对质疑文章的证据评价 文章最强的部分是可复现的仓库历史;较弱的推理包括: - 用学校声望推断 Offer 不可能; - 把所有数据工作一概排除在 AI 基础设施之外; - 与简历核验无关的性别、人格、税务与道德修辞; - 号召迫使当事人退网。 证据报告应排除这些内容,只把文章作为带归因的指控与链接来源。 ## 样例结论 > 公开证据支持真实实习、正式 Apache Committer 角色和部分有意义的开源贡献;现有证据不足以支持 DeerFlow 核心作者及主导 1.0 到 2.0 的完整强度。两组付费转化率与同屏用户数存在算术不一致。内部 AI 所有权、性能、Offer 与学历品牌主张在缺乏雇主或官方凭证时仍不可核验。证据支持“角色与范围存在重大膨胀风险”,不支持“所有经历均为虚构”的笼统结论。skills/asu-resume-audit-skill/references/education-branding-audit.md2.1 KBb8d2ec5cb93dupstream ↗
# 学历与中外合作项目表述审计 ## 不以学校声望作为证据 “双非”、排名、录取分数或社会声誉既不能证明也不能推翻学历。只核验正式教育事实以及简历如何呈现这些事实。 ## 建立院校关系图 每条教育经历分别记录: 1. 学籍院校:正式录取与注册所在学校; 2. 授课院校/校区:实际完成教学的地点; 3. 合作院校:参与联合、交换或课程合作的学校; 4. 项目性质:联合学院、合作专业、交换、访问、双学位、衔接项目或普通合作; 5. 学位授予方:出现在每张学位证书上的学校; 6. 实际获得凭证:学位、毕业证、项目证书或仅有参与证明。 ## 强证据 优先顺序: - 教育主管部门的官方项目名录; - 对应年份的大学项目页与招生简章; - 经授权提供的学籍/学位核验; - 正式成绩单或学历认证; - 项目措辞发生变化时的官方网页存档。 候选人主页与宣传手册的证明力低于正式凭证。 ## 措辞测试 ### 通常准确 ```text X 大学 · Y 联合学院|计算机本科|学位授予方:X 大学 X 大学|2025 年 9-12 月赴 Y 大学交换 X 与 Y 双学位项目|分别由 X、Y 授予学位 ``` ### 可能误导 ```text Y University|BSc in ... ``` 如果 Y 仅是合作方、交换接收方、课程提供方或品牌参与方,并未授予学位,该写法可能让读者误以为候选人正式就读于 Y。 ### 必查问题 - 海外合作方是否被视觉放大,而学籍院校被弱化? - 标题是否暗示在合作大学正式就读? - 是否省略 `联合`、`交换`、`访问`、`项目` 等限定词? - 展示的英文校名是否与正式学历凭证一致? - 是否把两所学校写成都授予学位,而实际上只有一方授予? ## 结论措辞 - `合作项目本身已证实,但当前写法可能扩大读者对学位授予方的理解。` - `合作关系真实;现有证据不能证明合作方授予了学位。` - `没有正式凭证或项目记录,暂不能形成结论。` 绝不能因为学校不属于名校就写“学历造假”。skills/asu-resume-audit-skill/references/evidence-methodology.md3.2 KB64a3a05200b8upstream ↗
# 证据方法论 ## 来源层级 每条主张都使用能够取得的最强来源。低层级来源可以提供线索,但不会因为谈论某个权威组织而自动获得该组织的可信度。 | 层级 | 来源类型 | 常见示例 | 能够证明什么 | |---|---|---|---| | A | 官方且同期的直接记录 | 雇佣核验、学历记录、官方项目名册、已合并 PR、正式任命、官方获奖名单 | 正式角色、学历、已交付产物、任命、奖项 | | B | 可复现的公开一手记录 | 仓库历史、Release Note、第一方网页存档、当事人公开发言 | 公开说过或发生过什么;部分情况下可证明贡献范围 | | C | 有署名的二手报道 | 带链接与可复现证据的署名文章 | 调查线索与上下文解释 | | D | 未核验陈述 | 匿名帖、无出处截图、评论区、情绪性表达 | 只能证明“有人这样说过”,不能证明内容为真 | ## 结论状态规则 ### `corroborated`(已证实) 独立的 A/B 级证据支持主张的关键措辞、角色与范围时使用。一个已合并 PR 只能证明该次具体贡献,不能自动证明“核心作者”。 ### `partially_corroborated`(部分证实) 事实锚点真实,但简历扩大了它。例如: - 正式拥有某开源项目 Committer 身份,但公开贡献主要集中在官网/前端,简历却暗示核心引擎深度; - 确有已合并 PR,但简历声称主导整个从 1.0 到 2.0 的重写; - 确有实习,但 `AI Infra owner` 的范围无法得到支持。 ### `contradicted`(存在直接矛盾) 必须存在可确定的冲突,例如: - 同屏分子、分母算不出展示百分比; - 官方学位授予方与简历声称的学位来源冲突; - 完整同期的官方项目名册与简历声称的排他性正式头衔冲突; - 权威时间戳证明所声称日期不可能成立。 ### `unsupported`(缺乏支持) 对公开主张做过合理范围检索后仍无足够证据,尤其适用于“最年轻”“第一”“核心”等最高级。报告必须写明检索边界。 ### `unverifiable`(暂不可核验) 只有私有雇佣、薪酬、支付或内部项目记录才能判定时使用。没有公开证据不等于内部主张为假。 ## 置信度 - `high`:结论来自直接证据或确定性算术。 - `medium`:多个独立来源趋同,但仍可能存在内部或历史例外。 - `low`:主要依靠推断,或关键来源不完整。 ## 归因与引用 - 对争议性陈述使用归因:`文章作者指控`、`当事人自述`、`截图显示`。 - 只引用识别主张所需的最短文字。 - 保存原始日期、URL 与访问日期。 - 区分“页面现在显示”与“相关时间点曾显示”。 - 使用网页存档时记录存档时间戳。 ## 反证检查 在形成不利结论前逐项检查: 1. 是否存在支持部分主张的已合并贡献或正式头衔? 2. 内部贡献是否可能被 squash、镜像或使用服务账号提交? 3. 官方名单是完整名册还是示例名单? 4. 争议词是否可能存在合理的宽泛含义? 5. 普通读者是否可能用比审计者更窄的方式理解该措辞? 所有限制都必须写入报告。skills/asu-resume-audit-skill/references/inflation-patterns.md3.2 KB6c6ab771677eupstream ↗
# 简历夸大模式防御分类 这些模式用于组织证据与提出核验问题,是待验证假设,不是先验定罪。 ## 1. 角色升格 `参与者 / Contributor -> Maintainer -> 核心作者 -> Owner / Lead` 检查正式任命、持续贡献范围、设计决策权、评审责任与 Release 责任。 ## 2. 全系统叙事 / 范围漂白 个人只做了 UI、文档、数据清洗或胶水逻辑,却用整个系统的架构链描述自己。应把“候选人实际产物”与“候选人能够解释的系统架构”分开。 ## 3. 品牌强关联 把公司名、知名仓库、Star 数、投资机构或知名合作者放在显眼位置,让其声誉迁移到个人。必须区分“与该品牌有关联”和“对结果作出多少贡献”。 ## 4. 可见度耕作 在多个知名项目中主动寻找高可见、低风险的官网、文档、注释、前端或配置任务。该行为本身可以是正当的作品集建设;只有在后续被描述为深度核心所有权时才构成误导信号。 审计时统计: - 提交数量、合并数量与未合并数量; - 代码区域分布; - 是否集中在上线可见但架构影响有限的区域; - 简历是否省略了这些边界。 ## 5. 社交邻接转技术授权 与 Maintainer 互动、获得点赞、被回复、共同出现在 Issue 或参加活动,被当作技术权威或联合开发的证据。社交与建立信任并非不当;应核验正式角色与交付产物,不能把“认识谁”当作“实现了什么”。 ## 6. 团队结果个人化 项目流量、排名、收入、效率或融资结果被展示,却不说明个人增量贡献。要求给出归因机制、时间窗口和个人负责链路。 ## 7. 伪精确 看似精确的百分比和计数缺少分母、基线、时间窗口、样本或数据来源。 ## 8. 分母切换 展示一个宽口径用户总数,却用更窄漏斗阶段作百分比分母。必须用同屏数字重算,再要求给出真正分母。 ## 9. 没有比较全集的最高级 `最年轻`、`首个`、`No.1`、`Top`、`核心` 没有公开比较集合或定义。 ## 10. 自证循环 / 多平台重复 同一个自授角色在多个自有账号重复发布,搜索结果看起来像多方证实。审计要统计独立来源,而不是页面数量。 ## 11. 合作院校替代学籍 把海外合作方或联合项目品牌写成主要大学身份,却未准确说明学籍与学位授予方。 ## 12. 时间压缩 短期或重叠经历中出现多个“0 到 1”“Owner”“跨部门主导”。结合工件、团队证据与可行工时核验,不能仅凭目标宏大直接判假。 ## 13. 修正传播不对称 最强主张被广泛传播,而删除、限定或更正只触达少数人。只有原文与更正都得到保存时才能纳入报告。 ## 14. 术语邻接 因为工作发生在 AI 团队、数据管线或模型产品附近,就把任何前端、数据或运营任务统称为 `AI Infra`。应分别记录团队背景、任务对象、个人技术栈和基础设施责任。 ## 报告模板 每个模式都写明: - 观察到的信号; - 涉及主张; - 支持证据; - 反向证据或善意解释; - 置信度; - 哪些材料能够证实或推翻该模式。skills/asu-resume-audit-skill/references/open-source-audit.md3.5 KBc1040e17168eupstream ↗
# 开源贡献与角色审计 ## 四个必须分开的问题 1. 当事人是否参与过? 2. 哪些工作被提出、合并、发布或持续维护? 3. 获得了什么正式角色? 4. 简历措辞是否准确描述了工作的深度与广度? ## GitHub 取证清单 - `author:<login> repo:<owner>/<repo> is:pr` 搜索结果; - PR 状态:已合并、关闭未合并、进行中、重复或被替代; - Commit 与改动文件; - 代码区域:核心运行时、独立功能、前端、官网、文档、测试、配置、仅注释; - Review 评论、Request Changes 与最终合并人; - 变更是否进入 Release; - 只提 Issue 还是交付实现; - README 致谢、CODEOWNERS、Team 文件、Maintainer 名单与组织成员关系; - 提交日期是否落在声称的版本或架构阶段。 常用搜索: ```text https://github.com/<owner>/<repo>/pulls?q=is%3Apr+author%3A<login> https://github.com/search?q=org%3A<org>+author%3A<login>+is%3Apr&type=pullrequests ``` 使用 GitHub API 时保存查询语句和原始结果数;绝不能把 `closed` 当成 `merged`。 ## 贡献深度分级 | 级别 | 常见证据 | 简历安全措辞 | |---|---|---| | 0 - 仅可见 | Fork、Star、评论、无实现的 Issue | 关注/讨论该项目 | | 1 - 边缘贡献 | 文档、错字、官网、小 UI、配置 | 贡献文档/官网/UI 修改 | | 2 - 有边界实现 | 已合并缺陷修复或功能,责任范围清晰 | 实现/修复具体组件 | | 3 - 持续子系统所有权 | 多次合并、评审、设计决策、Release | 维护/负责具体子系统 | | 4 - 项目领导 | 正式核心角色,加上跨模块架构与发布责任 | Maintainer/核心作者/项目负责人 | 正式头衔与贡献深度可能不完全相同。报告应同时保留“头衔真实”和“公开技术范围有限”两类事实。 ## 常见夸大信号 ### 品牌或 Star 迁移 简历把项目 Star、排名或公司品牌放在个人经历旁。除非能够证明个人造成了其中的实质增量,否则 Star 只能作为项目背景。 ### 贡献数量漂白 GitHub Contribution 可能包含 Issue、评论、Fork Commit 与未合并 PR。必须单独统计已合并工件。 ### 官网到核心引擎的歧义 官网、文档和 Developer Experience 都是正当贡献。误导发生在简历省略代码区域,让读者推断为核心引擎或算法所有权时。 ### 主动合 PR 与角色升格 主动寻找可贡献任务、与 Maintainer 互动、修复前端/官网问题、通过持续贡献获得社区信任都可以是正当成长路线。审计不判断动机,只检查后续措辞: - 正式角色是否确实授予; - 角色适用于整个项目还是网站/子项目; - 简历是否把可见度高的外围工作升级为核心技术深度; - 社交互动是否被当作无法支持的工作证明。 不能把社交能力本身描述为欺诈。潜在问题在于后续如何描述角色和贡献。 ### 全系统架构叙事 候选人可能准确理解整个项目,却只实现其中很小一部分。要求其给出本人负责的设计文档、PR、Commit、测试与线上事件。 ## 公平性检查 - 内部公司贡献可能被 squash 或由服务账号提交; - Maintainer 权限不一定公开; - README 的“关键贡献者”列表可能只是示例而非穷举; - 关闭未合并的 PR 仍可显示努力和技术思考,但不能证明已上线影响。 存在这些例外时使用“与公开证据存在张力”,不要写绝对否定。skills/asu-resume-audit-skill/references/report-schema.md4.5 KB765f42dee089upstream ↗
# 审计报告数据格式 渲染器接收 UTF-8 JSON,未知字段会被忽略。 ```json { "metadata": { "title": "简历主张真实性审计", "subject": "公开资料对象或匿名候选人", "generated_at": "2026-08-13T12:00:00+08:00", "scope": "公开来源简历主张审计", "analyst": "Codex", "disclaimer": "结论描述现有证据,不评价人格,也不构成法律认定。" }, "summary": { "conclusion": "一段校准后的总括结论。", "highest_risk": ["C-001", "C-004"], "limitations": ["无法取得内部雇佣记录"] }, "sources": [ { "id": "S-001", "title": "官方项目 README", "url": "https://example.com", "type": "official_record", "publisher": "项目组织", "accessed_at": "2026-08-13", "reliability": "A", "notes": "同期公开记录" } ], "claims": [ { "id": "C-001", "category": "open_source_role", "resume_text": "简历中的短句原文", "normalized_claim": "被核验的原子化命题", "status": "partially_corroborated", "risk": "high", "confidence": "high", "evidence_for": ["S-002"], "evidence_against": ["S-001"], "metric": { "numerator": 200, "denominator": 2000, "displayed_percent": 10 }, "analysis": "说明真实事实锚点、扩大部分与最终结论。", "verification_steps": ["要求提供正式 Maintainer 任命记录"], "page_break_before": false } ], "patterns": [ { "name": "角色升格", "description": "Contributor 活动被描述为核心作者身份。", "signals": ["核心作者措辞", "已合并工件有限"], "claim_ids": ["C-001"], "counter_evidence": "存在一个有意义的已合并 PR。", "confidence": "medium", "page_break_before": false } ], "timeline": [ { "date": "2026-03-11", "title": "PR 合并", "detail": "一个有明确边界的前端修复被合并。", "source_ids": ["S-003"] } ], "next_steps": [ "要求提供声称子系统对应的设计文档与代码所有权记录。" ] } ``` ## 必填字段 - `metadata.title`、`metadata.subject`、`metadata.generated_at`; - 至少一个 `source`; - 至少一个 `claim`; - 每条 Claim 必须包含 `id`、`normalized_claim`、`status`、`risk`、`confidence`、`analysis`; - 所有引用的 Source ID 必须存在; - Pattern 中引用的 Claim ID 必须存在。 证据来源的 `url` 为可选字段;提供时必须是完整的 `http` 或 `https` URL。渲染器会忽略其他协议和相对路径,避免在自包含报告中生成可执行或本地链接。 允许的状态: ```text corroborated partially_corroborated contradicted unsupported unverifiable ``` 允许的风险: ```text low medium high critical ``` 允许的置信度: ```text low medium high ``` 字段标识保持英文,所有用户可见内容使用中文。 Claim 可选使用 `metric` 对象记录明确的比例口径。只有同时提供 `numerator`、`denominator` 和 `displayed_percent` 时,校验器才会按 `numerator / denominator * 100` 重算,并允许 0.5 个百分点的舍入误差。缺少任一字段时不会推断;分母为零或字段不可转换为数字会导致校验失败。 同一个 `metric` 对象还可选记录明确的变化口径。只有同时提供 `baseline`、`result`、`change_type` 和 `displayed_change` 时,校验器才会重算;缺少任一字段时不会推断。`change_type` 可以是 `relative_percent` 或 `percentage_points`。 ```json { "baseline": 100, "result": 125, "change_type": "relative_percent", "displayed_change": 25 } ``` `relative_percent` 按 `(result - baseline) / baseline * 100` 计算;其 baseline 不能为零。百分比指标使用 `percentage_points` 表示百分点变化: ```json { "baseline": 60, "result": 70, "change_type": "percentage_points", "displayed_change": 10 } ``` 上例表示从 60% 到 70% 增加 10 个百分点,而不是增加 10% 的相对变化;相对变化约为 16.67%。两种变化校验与比例校验一样允许 0.5 的绝对舍入误差。 通过或失败的比例或变化校验仅说明已提供数字之间的关系,不能单独证明候选人存在虚假陈述。可能存在未披露分母或口径时,应在 Claim 的分析和报告限制中说明。 `claims[]` 与 `patterns[]` 可选使用 `page_break_before: true` 控制长报告的 A4 分页;它只影响 PDF,不改变 HTML 内容。skills/asu-resume-audit-skill/assets/report_template.html18 KBf8e9bc045b3fupstream ↗
<!doctype html> <html lang="zh-CN"> <head> <meta charset="utf-8"> <meta name="viewport" content="width=device-width, initial-scale=1"> <title>简历真实性审计报告</title> <style> :root { --ivory: #FAF9F5; /* page background */ --slate: #141413; /* primary text */ --clay: #D97757; /* warm accent - CTAs, danger, highlights */ --oat: #E3DACC; /* secondary fill - chips, soft panels */ --olive: #788C5D; /* success, positive deltas */ --gray-100: #F0EEE6; /* subtle background */ --gray-300: #D1CFC5; /* borders, dividers */ --gray-500: #87867F; /* meta text, captions */ --gray-700: #3D3D3A; /* secondary body text */ --white: #FFFFFF; /* card surface */ --serif: ui-serif, Georgia, "Songti SC", serif; --sans: system-ui, -apple-system, "PingFang SC", "Microsoft YaHei", sans-serif; --mono: ui-monospace, "SF Mono", Menlo, monospace; --radius-panel: 12px; --radius-row: 8px; --border: 1.5px solid var(--gray-300); --rust: #B04A3F; --warning: #C78E3F; --info: #5C7CA3; } * { box-sizing: border-box; } body { margin: 0; padding: 56px 24px 96px; background: var(--ivory); color: var(--slate); font-family: var(--sans); font-size: 15px; line-height: 1.55; -webkit-font-smoothing: antialiased; } .page { max-width: 980px; margin: 0 auto; } header { margin-bottom: 44px; } .eyebrow, .label, .chip { font-family: var(--mono); font-size: 11px; letter-spacing: .06em; text-transform: uppercase; } .eyebrow { color: var(--clay); margin-bottom: 10px; } h1, h2, h3 { font-family: var(--serif); font-weight: 500; letter-spacing: -.01em; } h1 { font-size: clamp(32px, 6vw, 46px); line-height: 1.12; margin: 0 0 14px; } h2 { font-size: 27px; margin: 0 0 8px; } h3 { font-size: 20px; margin: 0 0 8px; } .subtitle { color: var(--gray-700); font-size: 16px; max-width: 760px; } .meta { color: var(--gray-500); font-size: 13px; margin-top: 16px; display: flex; flex-wrap: wrap; gap: 8px 16px; } section { margin: 0 0 58px; } .section-head { border-bottom: 1px solid var(--gray-300); padding-bottom: 12px; margin-bottom: 22px; } .section-note { color: var(--gray-500); font-size: 13px; } .summary-grid { display: grid; grid-template-columns: repeat(5, minmax(0, 1fr)); gap: 12px; } .card, .panel { background: var(--white); border: var(--border); border-radius: var(--radius-panel); } .card { padding: 18px; } .metric-value { font-family: var(--serif); font-size: 38px; line-height: 1; margin-bottom: 8px; } .metric-label { color: var(--gray-500); font-size: 12px; } .status-corroborated { border-left: 4px solid var(--olive); } .status-partially_corroborated { border-left: 4px solid var(--info); } .status-contradicted { border-left: 4px solid var(--rust); } .status-unsupported { border-left: 4px solid var(--warning); } .status-unverifiable { border-left: 4px solid var(--gray-500); } .callout { padding: 22px 24px; border-left: 4px solid var(--clay); background: var(--white); border-radius: var(--radius-panel); } .callout p { margin: 0; color: var(--gray-700); font-size: 16px; } .limitations { margin-top: 14px; color: var(--gray-700); } .limitations li { margin: 6px 0; } .flow { padding: 24px; overflow-x: auto; } .flow svg { min-width: 760px; width: 100%; height: auto; display: block; } .table-wrap { overflow-x: auto; border: var(--border); border-radius: var(--radius-panel); background: var(--white); } table { border-collapse: separate; border-spacing: 0; width: 100%; min-width: 850px; } th { text-align: left; background: var(--gray-100); color: var(--gray-500); padding: 12px 14px; border-bottom: 1px solid var(--gray-300); font-family: var(--mono); font-size: 11px; letter-spacing: .04em; } td { padding: 14px; border-bottom: 1px solid var(--gray-100); vertical-align: top; font-size: 13px; } tr:last-child td { border-bottom: 0; } .claim-id { font-family: var(--mono); color: var(--gray-500); white-space: nowrap; } .badge, .risk-pill { display: inline-flex; align-items: center; border-radius: 999px; padding: 4px 8px; font-family: var(--mono); font-size: 10px; white-space: nowrap; } .badge { background: var(--gray-100); } .badge.corroborated { color: #52663D; background: rgba(120,140,93,.14); } .badge.partially_corroborated { color: #3D5E7C; background: rgba(92,124,163,.14); } .badge.contradicted { color: var(--rust); background: rgba(176,74,63,.10); } .badge.unsupported { color: #8B6328; background: rgba(199,142,63,.14); } .badge.unverifiable { color: var(--gray-700); background: var(--gray-100); } .risk-pill { border: 1px solid var(--gray-300); color: var(--gray-700); } .risk-pill.critical { border-color: var(--rust); color: var(--rust); } .risk-pill.high { border-color: var(--clay); color: var(--clay); } .pattern-grid { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 16px; } .pattern { padding: 20px; } .pattern p { color: var(--gray-700); margin: 0 0 12px; } .pattern ul { margin: 0; padding-left: 20px; color: var(--gray-700); } .counter { margin-top: 14px; border-left: 3px solid var(--olive); padding: 8px 0 8px 12px; color: var(--gray-700); font-size: 13px; } .timeline { position: relative; margin-left: 8px; } .timeline:before { content: ""; position: absolute; left: 5px; top: 8px; bottom: 8px; width: 1px; background: var(--gray-300); } .event { position: relative; padding: 0 0 24px 30px; } .event:before { content: ""; position: absolute; left: 0; top: 7px; width: 11px; height: 11px; border-radius: 50%; background: var(--clay); border: 3px solid var(--ivory); } .event-date { font-family: var(--mono); font-size: 11px; color: var(--gray-500); } .event-title { font-weight: 620; margin: 3px 0; } .event-detail { color: var(--gray-700); } .steps { display: grid; gap: 10px; counter-reset: steps; } .step { counter-increment: steps; display: grid; grid-template-columns: 30px 1fr; gap: 12px; align-items: start; background: var(--white); border: var(--border); border-radius: var(--radius-row); padding: 14px 16px; } .step:before { content: counter(steps); width: 26px; height: 26px; border-radius: 50%; display: grid; place-items: center; background: var(--slate); color: var(--white); font-family: var(--mono); font-size: 11px; } .source { padding: 14px 0; border-bottom: 1px solid var(--gray-300); } .source:last-child { border-bottom: 0; } .source-top { display: flex; justify-content: space-between; gap: 16px; } .source a { color: var(--clay); text-decoration: none; border-bottom: 1px dotted var(--clay); } .source-meta { color: var(--gray-500); font-family: var(--mono); font-size: 11px; margin-top: 5px; } .source-note { color: var(--gray-700); font-size: 13px; margin-top: 6px; } footer { margin-top: 72px; padding-top: 20px; border-top: 1px solid var(--gray-300); color: var(--gray-500); font-family: var(--mono); font-size: 11px; } .empty { color: var(--gray-500); font-style: italic; } @media (max-width: 780px) { body { padding: 32px 16px 72px; } .summary-grid { grid-template-columns: repeat(2, minmax(0, 1fr)); } .pattern-grid { grid-template-columns: 1fr; } .source-top { display: block; } } @media print { @page { size: A4; margin: 15mm; } body { padding: 0; background: #fff; font-size: 11px; } .page { max-width: none; } section { margin-bottom: 32px; break-inside: auto; } .card, .panel, .pattern, .step, .event { break-inside: avoid; } a { color: var(--slate) !important; border: 0 !important; word-break: break-all; } .flow { overflow: visible; } } </style> </head> <body> <div class="page" id="report"></div> <script id="report-data" type="application/json">__REPORT_JSON__</script> <script> (function () { "use strict"; const data = JSON.parse(document.getElementById("report-data").textContent); const root = document.getElementById("report"); const statuses = { corroborated: "已证实", partially_corroborated: "部分证实", contradicted: "存在直接矛盾", unsupported: "缺乏支持", unverifiable: "暂不可核验" }; const risks = { low: "低", medium: "中", high: "高", critical: "严重" }; function el(tag, cls, text) { const node = document.createElement(tag); if (cls) node.className = cls; if (text !== undefined && text !== null) node.textContent = String(text); return node; } function safeUrl(value) { try { const url = new URL(value); return ["http:", "https:"].includes(url.protocol) ? url.href : null; } catch (_) { return null; } } function section(title, note) { const s = el("section"); const head = el("div", "section-head"); head.appendChild(el("h2", "", title)); if (note) head.appendChild(el("div", "section-note", note)); s.appendChild(head); return s; } function appendList(parent, items, cls) { if (!items || !items.length) return; const ul = el("ul", cls || ""); items.forEach(item => ul.appendChild(el("li", "", item))); parent.appendChild(ul); } const m = data.metadata || {}; const summary = data.summary || {}; const claims = data.claims || []; const sources = data.sources || []; document.title = (m.title || "简历真实性审计报告") + " - " + (m.subject || ""); const header = el("header"); header.appendChild(el("div", "eyebrow", "证据优先 / 简历主张审计")); header.appendChild(el("h1", "", m.title || "简历真实性审计报告")); header.appendChild(el("div", "subtitle", m.subject || "未命名审计对象")); const meta = el("div", "meta"); ["范围:" + (m.scope || "未说明"), "生成时间:" + (m.generated_at || "未说明"), "分析者:" + (m.analyst || "未说明")].forEach(x => meta.appendChild(el("span", "", x))); header.appendChild(meta); root.appendChild(header); const counts = {}; Object.keys(statuses).forEach(k => counts[k] = claims.filter(c => c.status === k).length); const overview = section("证据结论总览", "每个数字代表一条原子化简历主张,而不是对人的评分。 "); const grid = el("div", "summary-grid"); Object.keys(statuses).forEach(k => { const c = el("div", "card status-" + k); c.appendChild(el("div", "metric-value", counts[k])); c.appendChild(el("div", "metric-label", statuses[k])); grid.appendChild(c); }); overview.appendChild(grid); const callout = el("div", "callout"); callout.style.marginTop = "18px"; callout.appendChild(el("p", "", summary.conclusion || "尚未形成结论。")); overview.appendChild(callout); if (m.disclaimer) overview.appendChild(el("div", "section-note", m.disclaimer)); if (summary.limitations && summary.limitations.length) { const lim = el("div", "limitations"); lim.appendChild(el("strong", "", "已知限制")); appendList(lim, summary.limitations); overview.appendChild(lim); } root.appendChild(overview); const flow = section("审计流程", "从简历文本到可复核结论,所有不确定性都保留在证据账本中。 "); const flowPanel = el("div", "panel flow"); flowPanel.innerHTML = '<svg viewBox="0 0 880 150" role="img" aria-label="审计流程:解析简历、拆分主张、检索证据、交叉核验、生成报告">' + '<defs><marker id="arrow" markerWidth="10" markerHeight="10" refX="8" refY="3" orient="auto"><path d="M0,0 L0,6 L9,3 z" fill="#87867F"/></marker></defs>' + '<g font-family="system-ui, sans-serif" text-anchor="middle">' + '<g><rect x="10" y="36" width="140" height="70" rx="12" fill="#FFFFFF" stroke="#D1CFC5" stroke-width="1.5"/><text x="80" y="66" font-size="14" fill="#141413">解析简历</text><text x="80" y="87" font-size="11" fill="#87867F">文本 / 图片 / PDF</text></g>' + '<g><rect x="190" y="36" width="140" height="70" rx="12" fill="#FFFFFF" stroke="#D1CFC5" stroke-width="1.5"/><text x="260" y="66" font-size="14" fill="#141413">拆分原子主张</text><text x="260" y="87" font-size="11" fill="#87867F">角色 / 范围 / 指标</text></g>' + '<g><rect x="370" y="36" width="140" height="70" rx="12" fill="#FFFFFF" stroke="#D1CFC5" stroke-width="1.5"/><text x="440" y="66" font-size="14" fill="#141413">检索证据</text><text x="440" y="87" font-size="11" fill="#87867F">官方 / GitHub / 档案</text></g>' + '<g><rect x="550" y="36" width="140" height="70" rx="12" fill="#FFFFFF" stroke="#D1CFC5" stroke-width="1.5"/><text x="620" y="66" font-size="14" fill="#141413">交叉核验</text><text x="620" y="87" font-size="11" fill="#87867F">支持 / 矛盾 / 缺口</text></g>' + '<g><rect x="730" y="36" width="140" height="70" rx="12" fill="#E3DACC" stroke="#D97757" stroke-width="1.5"/><text x="800" y="66" font-size="14" fill="#141413">生成报告</text><text x="800" y="87" font-size="11" fill="#3D3D3A">JSON / HTML / PDF</text></g>' + '<path d="M150 71 L185 71" stroke="#87867F" marker-end="url(#arrow)"/><path d="M330 71 L365 71" stroke="#87867F" marker-end="url(#arrow)"/><path d="M510 71 L545 71" stroke="#87867F" marker-end="url(#arrow)"/><path d="M690 71 L725 71" stroke="#87867F" marker-end="url(#arrow)"/>' + '</g></svg>'; flow.appendChild(flowPanel); root.appendChild(flow); const highIds = new Set(summary.highest_risk || []); const highClaims = claims.filter(c => highIds.has(c.id) || c.risk === "critical"); const high = section("最高风险主张", "只展示最可能影响招聘判断、且证据差距最大的项目。 "); const highTable = el("div", "table-wrap"); const table = document.createElement("table"); table.innerHTML = "<thead><tr><th>编号</th><th>简历主张</th><th>结论</th><th>风险</th><th>核心分析</th></tr></thead>"; const tbody = document.createElement("tbody"); (highClaims.length ? highClaims : claims.slice(0, 5)).forEach(c => { const tr = document.createElement("tr"); const id = el("td", "claim-id", c.id); const text = el("td", "", c.resume_text || c.normalized_claim); const status = el("td"); status.appendChild(el("span", "badge " + c.status, statuses[c.status] || c.status)); const risk = el("td"); risk.appendChild(el("span", "risk-pill " + c.risk, risks[c.risk] || c.risk)); const analysis = el("td", "", c.analysis); [id, text, status, risk, analysis].forEach(x => tr.appendChild(x)); tbody.appendChild(tr); }); table.appendChild(tbody); highTable.appendChild(table); high.appendChild(highTable); root.appendChild(high); const matrix = section("完整主张与证据矩阵", "每一条结论均保留支持证据、反向证据和下一步核验方式。 "); const matrixWrap = el("div", "table-wrap"); const matrixTable = document.createElement("table"); matrixTable.innerHTML = "<thead><tr><th>编号</th><th>原子主张</th><th>状态</th><th>置信度</th><th>证据</th><th>下一步</th></tr></thead>"; const matrixBody = document.createElement("tbody"); claims.forEach(c => { const tr = document.createElement("tr"); tr.appendChild(el("td", "claim-id", c.id)); tr.appendChild(el("td", "", c.normalized_claim)); const st = el("td"); st.appendChild(el("span", "badge " + c.status, statuses[c.status] || c.status)); tr.appendChild(st); tr.appendChild(el("td", "", c.confidence === "high" ? "高" : c.confidence === "medium" ? "中" : "低")); const ev = el("td"); ev.appendChild(el("div", "", "支持:" + ((c.evidence_for || []).join("、") || "无"))); ev.appendChild(el("div", "", "反向:" + ((c.evidence_against || []).join("、") || "无"))); tr.appendChild(ev); const next = el("td"); appendList(next, c.verification_steps || []); tr.appendChild(next); matrixBody.appendChild(tr); }); matrixTable.appendChild(matrixBody); matrixWrap.appendChild(matrixTable); matrix.appendChild(matrixWrap); root.appendChild(matrix); const patternSection = section("识别到的包装与夸大模式", "模式仅是对证据结构的描述;每项同时展示可能的合理解释。 "); const pg = el("div", "pattern-grid"); (data.patterns || []).forEach(p => { const card = el("article", "panel pattern"); card.appendChild(el("div", "eyebrow", "置信度 " + (p.confidence === "high" ? "高" : p.confidence === "medium" ? "中" : "低"))); card.appendChild(el("h3", "", p.name)); card.appendChild(el("p", "", p.description)); appendList(card, p.signals || []); if (p.counter_evidence) card.appendChild(el("div", "counter", "反向证据/合理解释:" + p.counter_evidence)); pg.appendChild(card); }); if (!pg.children.length) pg.appendChild(el("div", "empty", "未记录模式。")); patternSection.appendChild(pg); root.appendChild(patternSection); const time = section("时间线", "时间顺序用于识别角色跃迁、版本先后和可能的时间压缩。 "); const tl = el("div", "timeline"); (data.timeline || []).forEach(item => { const event = el("div", "event"); event.appendChild(el("div", "event-date", item.date)); event.appendChild(el("div", "event-title", item.title)); event.appendChild(el("div", "event-detail", item.detail)); tl.appendChild(event); }); time.appendChild(tl); root.appendChild(time); const nextSection = section("下一步核验", "把争议转化为可以回答的问题,而不是继续堆叠情绪。 "); const steps = el("div", "steps"); (data.next_steps || []).forEach(x => steps.appendChild(el("div", "step", x))); nextSection.appendChild(steps); root.appendChild(nextSection); const sourceSection = section("来源与证据等级", "A 为官方/直接记录;B 为一手公开自述;C/D 仅作为线索或指控来源。 "); sources.forEach(s => { const item = el("div", "source"); const top = el("div", "source-top"); const sourceUrl = safeUrl(s.url); const title = sourceUrl ? el("a", "", s.id + " · " + s.title) : el("strong", "", s.id + " · " + s.title); if (sourceUrl) { title.href = sourceUrl; title.target = "_blank"; title.rel = "noopener noreferrer"; } top.appendChild(title); top.appendChild(el("span", "badge", "等级 " + (s.reliability || "?"))); item.appendChild(top); item.appendChild(el("div", "source-meta", [s.type, s.publisher, s.accessed_at].filter(Boolean).join(" · "))); if (s.notes) item.appendChild(el("div", "source-note", s.notes)); sourceSection.appendChild(item); }); root.appendChild(sourceSection); const footer = el("footer", "", "由 asu-resume-audit-skill 生成 · 结论仅反映报告时间点的证据快照 · " + (m.generated_at || "")); root.appendChild(footer); })(); </script> </body> </html>skills/asu-resume-audit-skill/agents/openai.yaml338 B3b53729be580upstream ↗
interface: display_name: "ASU 简历质检" short_description: "核验角色膨胀、开源贡献、学历表述与指标口径并输出报告" brand_color: "#D97757" default_prompt: "使用 $asu-resume-audit-skill 对这份简历逐条核验,并生成中文 HTML/PDF 证据报告。" policy: allow_implicit_invocation: trueskills/asu-resume-audit-skill/evals/evals.json1.5 KB5ed9abd4fc6eupstream ↗
{ "skill_name": "asu-resume-audit-skill", "evals": [ { "id": 1, "prompt": "我有一份候选人简历 PDF,里面写自己是某 5 万 star 开源项目的核心作者。请核验 GitHub 贡献、正式角色、学校中外合作项目和所有业务指标,输出证据矩阵以及 HTML/PDF 报告。", "expected_output": "拆分原子主张;区分已合并、未合并、官网/文档与核心工作;审计学历凭证的精确措辞;重新计算指标;生成带来源 JSON、自包含 HTML 和 PDF,并使用校准结论。", "files": [] }, { "id": 2, "prompt": "有人发帖说这个实习生伪造 offer、靠和 maintainer 搞关系拿 committer 后冒充核心开发。我把帖子截图和 GitHub 链接放在附件里了,帮我打假并做一份能公开发布的报告。", "expected_output": "把帖子当作待核验指控,只使用与任务有关的公开证据;说明社交本身不是不当行为;核验正式角色与个人工件;排除骚扰和隐私内容;没有直接证据时不直接认定欺诈。", "files": [] }, { "id": 3, "prompt": "只检查这份简历的数字和时间线:注册 2000、付费 200、转化率 38%,四个月内同时写了三个从零到一项目。不要上网,给我一份简短 HTML 和 PDF。", "expected_output": "计算出 10%;除非存在另一分母,否则把 38% 标记为内部不一致;把时间压缩列为信号而非定案;明确说明未联网的证据限制。", "files": [] } ] }README.mdupstream ↗
Contents are not inlined in the catalog. Open the upstream file.
Plain-text copy, no login: /skill-bundles/claycui828-asu-resume-skills-skills-asu-resume-audit-skill.md