Skip to content
View BiBoyang's full-sized avatar
💪
A true master is an eternal student.
💪
A true master is an eternal student.

Block or report BiBoyang

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
BiBoyang/README.md

Hi, I'm BiBoyang 👋🏻

Full-stack developer going all-in on AI agents — AI-native builder. Now focused on multi-platform agent application architecture, agent memory, and evaluation. Formerly iOS, with a focus on performance optimization. Open-source and writing enthusiast, maintaining 45 public source repositories — DeepSeek Harness (DSH) plugins, a personal agent-skill collection with its own integrity gate, native macOS apps, and a Rust playground where a WeChat bot and a file-based agent prototype evolve side by side. Every plugin and skill ships with CI; every skill carries its own declarative evals.

🌍 Where things live

▶ Start here

Two repos lead the account:

Repository What it gives you
dsh-eval-harness DSH 插件评测工具:YAML 用例驱动真实 agent 回归评测,baseline 对比 PASS/WARN/FAIL 门禁
dsh-im-bridge DSH 插件:把 DeepSeek Harness 桥接到 IM(v0.1 微信/iLink;钉钉/飞书/Telegram 预留),turn/approval 推送 + 远程批准/注入,持久去重/收敛分段/合并窗口

🤖 Agents, plugins & skills

Repository One-liner Language License
dsh-eval-harness Regression eval harness for DSH plugins: YAML cases, real agent runs, baseline-diff gates TypeScript —
dsh-im-bridge Bridge DeepSeek Harness into IM: push turns/approvals, approve or inject remotely TypeScript —
huohou 个人的 AI agent skill 合集——每个 skill 是一条练到成本能的工作纪律;SKILL.md + references/ 渐进加载,每 skill 内置 evals 与触发评测(P/R/F1) Python —
skill-quake Fault-injection experiments for Agent Skills + skill-guard: the mechanical integrity gate (foundation accepts/rejects; grading is for judges) Python Apache-2.0
skills 个人 Agent Skill 注册仓:skills.yaml 单源真理 + symlink 同步器 + CI 对每个 pin ref 跑 skill-guard 验收 Shell —

🔧 Upstream & community contributions

Every repo below is external to BiBoyang/*; every entry in the first table is a merged pull request, re-derived 2026-10-06 from this account's own pull-request history. The owner column names who owns the project, because the account name alone does not say whether that is a company or an individual.

Merged — 6 pull requests across 3 repositories:

Repository ★ Merged PRs 项目归属方
awesome-dsh-plugin 17,881 #12, #13 — 收录 dsh-im-bridge 与 dsh-eval-harness 进精选列表 awesome-dsh-plugin community org — the community catalog, no company behind it
Waza 7,111 #26 中文技术文写作规则、#82 check skill 新增「质疑路线而不仅是 diff」、#88 本地抓取层绕过系统代理——堵住隐私契约的漏洞 tw93 — individual maintainer
skill-up 493 #257 符号链接 skill 源导致零文件静默安装 + 修复 阿里巴巴 Alibaba — the official alibaba org

Open — 6 pull requests across 5 repositories (measured 2026-10-06, re-checked weekly):

Open is open: not merged, not claimed. Two of these sit in official company orgs — Anthropic and Microsoft — and they are listed here exactly because they are still pending, not in spite of it.

Repository ★ Open PR 项目归属方
skills 179,860 anthropics/skills#1949 — skill-creator quick_validate 拒绝空 name/description 并校验引用文件存在性 Anthropic — the official anthropics org
waku 1,567 egoist/waku#161 — 修复表格内拖选复制带出整行内容(Rust 命中判定从纯垂直距离改为二维包含) egoist — individual maintainer
eval-guide 138 microsoft/eval-guide#16 — 修复重建说明里的过期路径 + 两条 frontmatter 非法 YAML Microsoft — the official microsoft org
skill-up 493 #295 拒绝无任何可识别 matcher 的空规则断言、#281 validate --skill 内容完整性 linter、#265 截止杀掉进程后合成 session-result 产物 阿里巴巴 Alibaba — the official alibaba org

The biggest merged row is a catalog and an individual's project, and the table says so rather than letting the star count imply otherwise. The merged company row is Alibaba's skill-up — the eval engine this account's own skill-quake and skills repos build their gates around, contributed to upstream instead of only consuming it. Three merged pull requests in Waza also feed directly back into the writing and review disciplines that huohou ships as skills; the open rows to Anthropic and Microsoft extend the same line: integrity checks and correctness fixes, proposed where the ecosystem will feel them.

🍎 Apple platform

Repository One-liner
Cove A native macOS NAS media player (Swift, MIT)
NotchLauncher A macOS launcher that lives in the notch — hover to reveal, click to launch (Swift, MIT)
ForgeLoop Swift coding-agent project: layered architecture, streaming, tool execution, cancellation semantics (MIT)
ForgeLoopTUI Lightweight Swift terminal UI library for streaming AI transcripts with in-place updates and tool placeholders (MIT)
BBYDebugTool LogTool v0.1 — Objective-C logging utility, the account's oldest repo (last commit 2020-04)

🦀 Rust experiments

Repository One-liner
AMClaw 实验场:微信 iLink Bot(扫码登录、会话合并、记忆、熔断、日报/周报回传)+ 最小文件型 Agent 原型(Plan-aware ReAct、watchdog、LLM record/replay);破坏性演进是特性
AIvsAI Dual-AI terminal tool: Moonshot answers, DeepSeek reviews (MIT)

Note: the eight repos above lead the account by stars (ordered by ★, measured 2026-10-06). The remaining 30+ source repos are learning archives, forks-in-progress and small demos, named only in aggregate.

✍️ Writing

  • Blog — biboyang.github.io — AI agent 工程、流式协议故障注入、skill 完整性实验、CI 误诊纪实;写「我具体错在哪条命令上」的那种文章
  • 《自顶向下拆 Agent:从界面到地基》 — 博客系列「自顶向下拆 Agent」全部篇目集结的网页版小册,从界面现象一路拆到上下文与协议地基

🚴 Beyond code

  • 📚 Avid reader
  • 🎮 Gaming — League of Legends
  • 🚴🏻 Biking — Trek Marlin 5 👍🏻👍🏻👍🏻, Marin Four Corners 👍🏻👍🏻👍🏻👍🏻👍🏻

中文介绍

All-in AI Agent 的全栈开发者,AI Native 开发者,现在专攻多端 agent 应用架构、 agent 记忆与评测;曾经做 iOS 开发,主攻性能优化。 这个账号维护 45 个公开源码仓:DeepSeek Harness 插件(评测门禁、IM 桥接)、一套带机械完整性闸门 的个人 agent skill 合集(火候)、原生 macOS 应用,以及一个 Rust 实验场——微信 Bot 和文件型 Agent 原型在里面一起演化。向上游也交代码:6 条已合并的外部 PR(阿里 skill-up、tw93 的 Waza、DSH 社区精选列表),另有 6 条 open PR 悬在 5 个外部仓上 ——其中两条分别开在 Anthropic 官方 skills(179.9k★)和 Microsoft 官方 eval-guide, 逐条见上方 Upstream 一节。写作在博客 biboyang.github.io, 系列长文已集结成小册《自顶向下拆 Agent:从界面到地基》。上面的每个数字都带测量日期, 不复述第二份。


This page is re-measured, not remembered. Star counts, merged-pull-request counts and repository figures were re-measured on 2026-10-06 against the GitHub API and the account's public repository list, and each table says so. Anything that can't be measured is labeled instead of guessed.

Pinned Loading

  1. dsh-eval-harness dsh-eval-harness Public

    DSH 插件评测工具:YAML 用例驱动真实 agent 回归评测 + baseline 对比 PASS/WARN/FAIL 门禁|Regression eval harness for DeepSeek Harness plugins

    TypeScript 13 1

  2. tw93/Waza tw93/Waza Public

    🥷 Engineering habits you already know, turned into skills Claude can run.

    Python 7.2k 415

  3. alibaba/skill-up alibaba/skill-up Public

    An evaluation and evolution tool for Agent Skills.

    Go 1.1k 95

  4. deepseek-ai/deepseek-harness deepseek-ai/deepseek-harness Public

    DeepSeek Harness: Everything is a Plugin.

    TypeScript 245k 29.3k

  5. MoonshotAI/kimi-code MoonshotAI/kimi-code Public

    Kimi Code CLI — The Starting Point for Next-Gen Agents

    TypeScript 7.8k 1.3k

  6. egoist/waku egoist/waku Public

    ⚡ A native app for all your coding agents.

    Rust 1.6k 184