adversarial-review
Skill0 starsChanged 2 months agoMITView on GitHub
行途 XingTu 自媒体"多专家对抗审查"技能(豆包工作优化版)。对公众号封面/标题/正文/数据/方案 做多角色交叉对抗审查,消除单视角盲区。支持怀疑论者/优化者/审计员/过滤器四角色并行, Round 2 脚本交叉验证(不用 LLM 法官),仅保留高置信度发现。 触发词:对抗审查、交叉审查、多视角审查、审文章、审标题、审封面、审查方案、数据核查。 依据:RULES.md PUB-017(多专家对抗审查)+ 行途 XingTu 内容工厂五关 + 原版 adversarial-review。 版本:豆包工作优化版 v1.0(2026-09-03);原版归档 .backups/skills_原版归档_20260903/adversarial-review_原版归档(内部项目)
107 other projects use a version of this, 1 of them edited.
Adopted by
Most recently taken up by
- magnusrodseth/dotfilestheir own version, first seen Nov 16, 2026
- lennoxsaint/Skillstheir own version, first seen Nov 15, 2026
- fabioc-aloha/Alex_ACT_ONEtheir own version, first seen Nov 15, 2026
- samuelralak/How-I-Worktheir own version, first seen Nov 15, 2026
- baekenough/oh-my-customcodetheir own version, first seen Nov 14, 2026
- leinardi/adversarial-review-looptheir own version, first seen Nov 13, 2026
Used with
Projects with it also use rethink, root-cause-analysis and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
fabioespindulaSkillrethink
Reframe a plan, idea, or decision before the user commits to it. Use when the user asks to rethink, step back, pressure-test, gut-check, decide, right-size, simplify, think bigger about, or evaluate a concrete plan or decision in product, strategy, technical architecture, career, life, company direction, or personal priorities. This skill improves decision quality before action: clarify the real question, challenge assumptions, compare alternatives, calibrate ambition, classify the operating mode, surface risks, and recommend a next move. Do not use for broad future exploration without a concrete decision; that belongs to a think-big or research-style skill.
Also in 3 projects that use this
MicrosoftSkillroot-cause-analysis
Use when a Regional Retail Manager asks I'm seeing unusual inventory movement at our Northwest stores. Can you help me understand what's happening?
Also in 3 projects that use this
anasahmed07Skillbetter-auth-best-practices
Skill for integrating Better Auth - the comprehensive TypeScript authentication framework.
Also in 4 projects that use this
DockerSkillfix-issue
Fix a single GitHub issue end-to-end: triage, research, write the fix, review, and create a PR. Use when asked to fix an issue: "fix issue 1234", "resolve #500", "create a PR for issue 200".
Also in 3 projects that use this
MicrosoftSkillpull-request
Write concise, focused pull request descriptions for Wassette — at most three sentences on the what and why, with an explicit note on any breaking public-API change and how users adapt. Use when opening or editing a Wassette pull request.
Also in 3 projects that use this
juriwiensSkillcodex-review
Run code review via Codex CLI
Also in 3 projects that use this
- +2 more from microsoft
Other versions107 projects use a version of this.
Word for word · 0
None found.
Edited · 1
- xingtu1996/dsh-xingtu-skills★ 294% same
Wrote their own version · 106
- PrefectHQ/fastmcp★ 29ksame name
- jlowin/fastmcp★ 29ksame name
- mengxi-ream/read-frog★ 10ksame name
- epicenter-md/epicenter★ 4.8ksame name
- EpicenterHQ/epicenter★ 4.8ksame name
- basicmachines-co/basic-memory★ 4.6ksame name
- DanMcInerney/architect-loop★ 626same name
- deadlock-mod-manager/deadlock-mod-manager★ 584same name
- boraoztunc/skills★ 473same name
- syahiidkamil/Software-Engineer-AI-Agent-Atlas★ 404same name
96 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository139 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/caveman-learn/CLAUDE.mdCLAUDE.md
- ai-repo-engineering-layoutSkill
- ai-session-material-minerSkill
- ai-usage-reportSkill
- ardot-effect-fallbackSkill
- audio-friendly-adaptSkill
- automation-resurrectionSkill
- book2vido-videoSkill
- boss-autopilotSkill
- brand-icon-sourcingSkill
- canvas-designSkill
- cavecrewSkill
- caveman-commitSkill
- caveman-compressSkill
- caveman-discoverSkill
- caveman-evidence-reviewSkill
- caveman-exploreSkill
- caveman-helpSkill
- caveman-learnSkill
- caveman-manageSkill
- caveman-optimizeSkill
- caveman-reviewSkill
- caveman-setupSkill
- cavemanSkill
- caveman-statsSkill
- cc-switch-usage-auditSkill
- certification-submission-kitSkill
- comment-reply-playbookSkill
- concept-anchor-figureSkill
- conference-archiveSkill
- constitution-amendmentSkill
- content-format-auditSkill
- content-matrix-channel-auditSkill
- cross-repo-skill-curationSkill
- daily-judgment-briefSkill
- daily-reportSkill
- de-ai-flavorSkill
- desktop-ai-forensicsSkill
- discernment-nudgeSkill
- doc-translateSkill
- doubao-app-layer-archiveSkill
- effort-accountingSkill
- ego-browserSkill
- egress-allowlistSkill
- eval-skillsSkill
- evolve-skillsSkill
- exam-prep-pmpSkill
- fact-check-gateSkill
- fanout-mainline-ledgerSkill
- free-token-channel-auditSkill
- github-repo-gallerySkill
- git-identity-pinSkill
- git-push-in-sandboxSkill
- harness-governanceSkill
- harness-model-upgrade-auditSkill
- hidden-contract-auditSkill
- investigate-firstSkill
- ip-platform-fit-auditSkill
- lean-buildSkill
- The first 60 shown.
The file
Files
对抗性审查引擎(Adversarial Review Engine · 行途 XingTu 版)
来源:原版 adversarial-review(内部项目实践提炼)+ 行途 XingTu 内容工厂五关 + PUB-017 多专家对抗审查 适配日期:2026-09-03 核心理念:单一视角的审查有固有盲区。 多角色从对立视角独立审查 → 脚本交叉验证(不用 LLM 法官)→ 仅保留高置信度发现。
〇、Token 纪律(对齐省 token 系列)
铁律:对抗审查的 token 消耗必须符合省 token 哲学。并行提升效率,但不能让 token 爆炸。
| 原则 | 规则 |
| TKR-1 | 审查角色输出精简,发现列表不写散文 |
| TKR-2 | Round 2 交叉验证用脚本(grep/ls/diff/查库),不用 LLM 当法官(省一份法官的 token) |
| TKR-3 | 每个角色发现上限 7 条——太少=草率,太多=噪音浪费 token |
| TKR-4 | 领域专家(封面设计/数据/合规)按需激活——只在审查目标涉及对应领域时启用 |
| TKR-5 | 轻量审查(单篇文章)→ 3 角色并行。标准审查(系列/方案)→ 4 角色并行。重量审查(付费/商业决策)→ +领域专家 |
| TKR-6 | 审查结果直接写回被审文件的"审查记录"章节,不创建独立审查报告(除非用户要求) |
审查深度决策
| 级别 | 触发 | 角色数 | Round | Token 估算 |
| 轻量 | 单篇封面/标题 | 3 (Refuter + Optimizer + Auditor) | 并行 1 轮 | ~10K |
| 标准 | 单篇文章正文/数据 | 4 (全部) | 并行 1 轮 + 脚本交叉 | ~25K |
| 重量 | 系列规划/付费方案/合集策略 | 4 + 领域专家 | 并行 1 轮 + 专家审 + 脚本交叉 | ~50K |
一、行途 XingTu 自媒体四角色审阅模型
审查对象(封面 / 标题 / 正文 / 数据 / 方案)
│
┌──────────────┼──────────────┐
▼ ▼ ▼
怀疑论者 优化者 审计员
(Refuter) (Optimizer) (Auditor)
找到漏洞 提出改进 验证数据和一致性
"这结论站得住吗?" "还能怎么优化?" "数字对得上吗?"
│ │ │
└──────────────┼──────────────┘
▼
┌──────────────┐
│ 过滤器 │
│ (Filter) │
│ 误报拦截 │
└──────┬───────┘
▼
脚本交叉验证 → 置信度分级 → 写回| 角色 | 职责 | 问的核心问题 | 不看什么 |
| 怀疑论者 (Refuter) | 找到论证的薄弱点 | "这个结论的证据充分吗?""数字有来源吗?""遗漏了什么成本/反例?" | 不改建议、不优化格式 |
| 优化者 (Optimizer) | 找到可以更好的地方 | "标题能更抓人吗?""表述能更清晰吗?""有更省的写法吗?" | 不找逻辑错误 |
| 审计员 (Auditor) | 验证数据和引用 + 精确位置定位 | "封面数字=正文数字吗?""来源真实存在吗?""口径一致吗?" | 不做价值判断 |
| 过滤器 (Filter) | 误报拦截——过滤低质量发现,只保留高置信度 | "这个发现是真问题还是噪音?""置信度够高吗?" | 不新增发现、不修改发现内容 |
审计员位置定位规则
解决"AI 评论的行号对不上实际内容"的经典问题。文章审查用章节/字段定位,封面用数字+文案定位。
❌ 旧方式(位置漂移):
- 位置: S05 封面 第 3 行
- 问题: 数字口径不一致
✅ 新方式(字段定位):
- 位置: 封面 S05 big 字段(HTML: <span class="big">)
- 内容: "93.6%" → 正文主数字是 "8880万/30天"
- 问题: 封面与正文口径不一致,读者会混淆二、两轮流程
Round 1:独立并行审查(3-4 角色并行,精简输出)
每个角色收到相同的审查对象 + 角色定义,独立产出发现列表。
输出格式(每个发现):
### [角色] 发现 #N: {一句话摘要}
- 位置: {封面/正文§章节/标题}
- 严重度: 🔴 严重 / 🟡 需关注 / 🟢 建议
- 证据: {具体引用或数据}
- 理由: {为什么这是问题}Round 2:脚本交叉验证 + Filter 误报过滤
Round 1: Refuter + Optimizer + Auditor (+专家) 并行审查
↓
Round 2a: 脚本交叉验证(去重 + 置信度分级)
↓
Round 2b: Filter 误报过滤(拦截低质量发现)置信度分级:
✅ 交叉验证 = ≥2 个角色独立报告了同一发现
🔹 单独 = 仅 1 个角色报告Filter 角色职责:
- 对"单独"置信度的发现逐条检查——是否真正有价值?
- 过滤标准:没有具体证据的 / 纯主观偏好的 / 表述模糊的 → 标记为
[已过滤] - 不新增发现、不修改发现内容——只做减法
- 目标:拦截 30-50% 的低质量发现
三、行途 XingTu 内容审查角色要点(豆包工作优化版新增)
怀疑论者检查要点(文章/封面)
- 标题是否 ≤13 字且核心词前置(PUB-014)?还是堆了纯数据钩子?
- 数字是否有来源?是"实测/官方/估算"哪种口径?
- 有没有遗漏的代价/反例?(如:压缩过头会丢关键报错)
- 数据脱敏是否符合 PUB-018(公司数字脱敏到量级)?
- 有没有编造数字?(对应用户"凭印象硬编是大忌"红线)
优化者检查要点
- 标题/封面 caption 读者能秒懂吗?(SignalDistilled:数字第一语言)
- 有没有更省 token 的写法?(对齐省 token 系列母题)
- 金句锚点(思想定调)够不够打动人?
- 实操命令读者能直接抄吗?(每篇至少 1 段)
审计员检查要点(必须用工具验证)
- 封面数字 = 正文主数字?(跑
check_cover_consistency.py) - 数据来源真实吗?(cc-switch 库 / 内部 AI 平台 API / 素材矿报告)
- 官方口径 vs 实测口径是否分开标注?(对齐 S05 三数口径对比)
- 文件路径/引用真实存在吗?(
ls/test -f) - 日期/版本号准确吗?
领域专家(按需激活)
| 场景 | 专家 | 检查什么 |
| 封面/视觉 | 封面设计专家 | SignalDistilled 哲学、行途蓝 VI、900×383 |
| 数据/账单 | 数据审计专家 | 数字口径、缓存命中率、成本核算 |
| 合规/脱敏 | 合规专家 | PUB-018 脱敏、雇主信息、隐私暴露面 |
四、执行指令(给 AI Agent 的标准提示词)
你是审查对象 [封面/文章/方案] 的对抗性审查员。请以 [角色] 的视角检查。
角色定义:[从 §一 角色表复制]
检查要点:[从 §三 角色要点复制]
审查范围:[指定封面HTML/文章md/方案]
输出要求:
- 每个发现:位置 + 严重度 + 证据 + 理由
- 精简输出(TKR-1):不写"看起来不错"等无信息量文字
- 发现数量:3-7 条(TKR-3)五、审查报告模板
# [审查对象] 对抗性审查报告
> Round 1: 3-4 角色并行 | Round 2: 脚本交叉 | 修复: N 项 | 接受: M 项
## 交叉验证发现(≥2 角色独立发现)
| # | 发现 | 严重度 | 报告者 |
|---|------|:--:|------|
| 1 | xxx | 🔴 | Refuter + Auditor |
## 全部发现
| # | 发现 | 角色 | 置信度 | 严重度 | 状态 |
|---|------|------|:--:|:--:|:--:|
| 1 | xxx | Refuter | ✅ 交叉验证 | 🔴 | ✅ 已修复 |
## 过滤统计
| 指标 | 值 |
|------|:--:|
| 总发现 | N |
| 过滤拦截 | M (M/N%) |
| 最终保留 | N-M |六、自检清单
| # | 检查项 |
| 1 | 是否用脚本(grep/check_cover_consistency/查库)验证过审计员的数据发现? |
| 2 | 🔴 严重发现(数字不一致/编造/脱敏漏洞)是否立即修复? |
| 3 | Round 2 是否用脚本而非 LLM 做交叉验证(TKR-2)? |
| 4 | 封面数字与正文主数字是否核对过口径? |
| 5 | 官方口径 vs 实测口径是否分开标注? |
| 6 | Filter 是否拦截了 ≥30% 的低质量"单独"发现? |
| 7 | 数据脱敏(PUB-018)是否过审? |
七、与其他 Skill 的关系
| Skill | 审查维度 | 触发方式 |
| adversarial-review | 逻辑 + 一致性 + 数据口径 + 脱敏 | 文章/封面/方案完成 / "审查" |
| de-ai-flavor | 去 AI 味 / 文案 | "去 AI 味" / 润色 |
| ai-usage-report | 数据核实 / 统计 | "核实数字" / "看账单" |
adversarial-review 是发布前的质量门禁——在 de-ai-flavor(去味)和内容定稿之后运行,数据问题交给 ai-usage-report 核实。
版本历史
| 日期 | 版本 | 变更 | 来源 |
| 2026-09-03 | v1.0 | 豆包工作优化版:适配行途 XingTu 公众号(封面/标题/正文/数据审查)+ PUB-017 对齐 + 领域专家表 + 数据核实联动 | 原版 某内部项目 + 某内部代码仓库 |
豆包工作优化版:适配行途 XingTu 自媒体场景,原版已归档
.backups/skills_原版归档_20260903/。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

