Skip to content
View chenhz01's full-sized avatar

Block or report chenhz01

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
chenhz01/README.md
Typing SVG

chenhz01

AI Agent 治理 × Skill 工程 做「把可审计质量推成 AI Agent 行业标准」的第一人。

AI Agent Governance Skill Engineering Zero Dependency arXiv-aligned Open to collaborate Email


恐惧失控是猜测;没有审计轨迹是事实。 我不卖安全感。我造的是让自进化 AI「可证明没跑偏」的机械门禁——证据不可篡改,标准公开可采用。

🧰 Toolbox — 55 purpose-built AI-engineering tools

Toolbox Score-first Zero dependency

A toolbox of 55 purpose-built tools across neglected AI-engineering niches — prompt-engineering harnesses, agent governance, code-quality gates, security specs. Every tool: pure stdlib, CLI-first, 40+ self-test cases, CI-verified. Entry point: zhengming-toolbox.

📏 我正在推动的行业质量标准(公开 · 可采用)

别的个人主页列项目,这里列标准。以下门禁全部开源、零依赖、可直接接入你的 Agent 工程:

# 标准门禁 一句话 采用入口
G1–G6 进化硬性拒绝护栏 自进化 AI 的六道审计关卡,跑偏即拒绝 EEL
P1–P5 发布突破五流程 从设计到真实验证 100% 覆盖的发表纪律 ai-agent-skill-breakthrough
D1–D5 研究对抗审计 覆盖率算术、分簇加总、引用完整性——抓"没在撒谎却在撒谎" adversarial-research-audit
M-Boot 记忆开机体检 抓约定静默丢失与注入稀释 agent-memory-doctor

🎯 把你的 Agent 项目接入以上任一门禁,开 Issue 标注 adoption——你就是我推动行业质量的一部分。

🛡️ 治理与审计 — 旗舰项目

项目 一句话
⭐ EEL 进化证据账本:为自进化 AI 造不可篡改的进化账本(6 篇 arXiv 论文 → G1–G6 硬性拒绝护栏)· 求合作
agency AI 智能体的纪律操作系统:对抗审计、发布门禁、37 身份对攻锻造、防篡改演化账本
adversarial-research-audit 机械门禁审计研究报告:覆盖率算术、分簇加总、引用完整性 · 零依赖
agent-memory-doctor AI 记忆与约定文件的开机体检:抓沉默的约定丢失与注入稀释 · 零依赖

🧩 Skill 工程 — 让 skill 库从"收藏夹"变成"生产力"

项目 一句话
skill-audit-kit skill 库健康检查:引用图谱、孤儿检测、冲突、语义近重复 · 零依赖 CLI · JSON+HTML 报告
answer-contract 四条规则止住 AI 注水:隐性需求补齐 / 结构化输出 / 三方案对比 / 假设熔断 · 单文件
zhengming-openai-plugins SKILL.md → OpenAI Codex 插件包转换器 · 纯标准库 · 强制官方 manifest 约束
ai-agent-skill-breakthrough 学→超→破→用:把开源 agent skill 集合真正用起来的方法论

📐 方法论与工程实践

ai-native-process-engine(AI 原生流程引擎 · 七层架构 + L0-L3 自治分级)· fork-to-mainrepo-playbook(衍生项目独立化方案)· trending-scout(Trending 对抗性审核)· PartnerGene(基因中心型 AI 商务建联)

🧪 行为层 conformance — 已被上游验证的轨迹

agent-memory-benchmark:行为层记忆质量基准——11 条机械可断言条款、零依赖 L1–L5 conformance checker、NMR 遗忘质量排行榜、AMTF 可移植 schema。

一个 90 秒的发现:对 mem0(v2.2.1)用 deepseek-chat(temperature=0)做行为探针——先说最爱 GPT-4o,再说已换成 DeepSeek-V3。记忆层正确存了带时间戳的更新,但查询"我最爱什么 LLM"返回自相矛盾的两条:被取代的旧记忆从未死去。该 FAIL 在 temperature=0 下确定性复现 N=2,证据逐字节一致。→ 完整报告 · 复现脚本 · 原始输出

上游信任锚(每条可点击验证):


🤝 正在寻找的合作

  • 谁:做 agent 平台 / skill 生态 / AI 治理的团队与独立开发者
  • 四种合作形态:
    1. 集成——审计引擎接入你的平台(EEL / adversarial-research-audit)
    2. 诊断——skill 库健康体检 + 修复路线(skill-audit-kit)
    3. 定制——SKILL.md→Codex 转换与 L3 配方层深度交付
    4. 中国区落地——你在广州的 on-the-ground partner:本地开发者社区、中文模型栈 conformance 跑批(DeepSeek / Qwen / GLM / Kimi)、中国市场准入的步步落地——步子随你的节奏,今天无需任何承诺
  • 怎么找:邮件 shanlun2029@outlook.com,或直接在对应仓库开 Issue(adoption / collab)

「进化是证据,证据不可篡改。」

如果你也在做 Agent 治理或 skill 生态——我们大概率该认识。 ⭐ Star 一个项目 · 📮 发封邮件

🌐 English

AI Agent Governance × Skill Engineering. The first person pushing auditable quality as the industry standard for AI Agents.

  • I build mechanical gates that let self-evolving AI prove it didn't drift — G1–G6 evolution guardrails, adversarial research audits, boot-time memory integrity checks.
  • I turn agent skill collections into production pipelines: health checks, contract rules, SKILL.md→Codex converters. All zero-dependency, all open source.
  • Behavior-layer conformance for agent memory (agent-memory-benchmark): 11 machine-assertable clauses, deterministic probes, raw outputs published. Sample finding: mem0 v2.2.1 keeps self-contradictory superseded facts alive at temperature 0 (deterministic, N=2) — full report & repro.
  • Merged upstream: my PR to TeleAI-UAGI/Awesome-Agent-Memory was merged to main by the maintainer with co-author credit.
  • Adopt any gate in your project — open an Issue tagged adoption and become part of the standard.
  • Eyeing the China market? I'm your on-the-ground partner in Guangzhou: local developer community, Chinese model-stack conformance (DeepSeek / Qwen / GLM / Kimi), and a China presence built step by step, at your pace.

Looking for: agent platform / skill ecosystem / AI governance teams. → shanlun2029@outlook.com

Pinned Loading

  1. adversarial-research-audit adversarial-research-audit Public

    Adversarial audit layer for agentic research: mechanical gates that catch reports which lie without lying — coverage arithmetic, cluster sums, filter provenance, citation integrity. Zero-dependency…

    Python 3 1

  2. agent-memory-doctor agent-memory-doctor Public

    Boot-time integrity check for AI agent memory & convention files — catch silent convention loss, injection dilution, broken rollback anchors. Zero dependencies, Python 3.8+.

    Python 3

  3. EEL EEL Public

    EEL — Evolution Evidence Ledger:为自进化 AI 造不可篡改的进化账本(求合作脱敏版)

    1

  4. zhengming-openai-plugins zhengming-openai-plugins Public

    Convert Agent Skills (SKILL.md) format to OpenAI Codex plugin bundles — stdlib-only, enforces official manifest hard constraints

    Python 1

  5. answer-contract answer-contract Public

    Four rules that stop your AI agent from padding, rambling, and shipping one mediocre answer: Implicit-Need Fill, Structured Output, 3-Option-Plus-Premium, Assumption-Circuit-Breaker. A single-file …

  6. skill-audit-kit skill-audit-kit Public archive

    Health check for an existing skill/rule library: reference graph, orphans, hubs, conflicts, and zero-dependency semantic near-duplicate detection. Offline CLI with JSON + HTML reports.

    JavaScript