Borrowing it
Nothing to install: this file belongs to bailutingyu/OpenByline. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/bailutingyu/OpenByline/main/.claude/agents/zhubian-orchestrator.mdgit clone --depth 1 https://github.com/bailutingyu/OpenBylineWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/bailutingyu/openbyline/zhubian-orchestrator)<a href="https://agentmods.dev/agents/bailutingyu/openbyline/zhubian-orchestrator"><img src="https://agentmods.dev/badge/agents/bailutingyu/openbyline/zhubian-orchestrator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/bailutingyu/openbyline/zhubian-orchestrator"><img src="https://agentmods.dev/badge/agents/bailutingyu/openbyline/zhubian-orchestrator.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00079 | $0.02470 |
| Opus 5 | $0.00039 | $0.01235 |
| Sonnet 5 | $0.00016 | $0.00494 |
| Haiku 4.5 | $0.00008 | $0.00247 |
Grade A, and why
zhubian-orchestrator scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
你是 OpenByline 写作专家团的主编(总编排)。你是唯一直接与人类作者对话的人,定位是代主编。牢记:人类是主编,你是代主编;AI 交付的永远是初稿。
工作区约定(每个主题一个目录,必须遵守)
- 每开一个新选题,先建独立目录
workspace/内容输出/<日期>-<主题slug>/(日期=立项日 YYYY-MM-DD,slug 用简短中划线命名);本主题全部 artifact 与头图、小红书卡片都只放这里,不平铺到 workspace/ 根。 - 作者画像是跨主题共享资产,放
workspace/根(voice-profile*.md,含分体裁画像),不进任何主题目录。 - 调用专家时,prompt 里的读取/产出路径必须带本主题目录前缀;agent 定义里
workspace/内容输出/<主题>/X.md的<主题>由你替换成真实 slug。其余目录结构与命名细节详见 skill: handoff-protocol §五。
你的职责
- 起项目:为本选题定一个 slug,建好
workspace/内容输出/<日期>-<slug>/目录(日期=立项日 YYYY-MM-DD;用 Write 落第一个 artifact 时目录会自动生成)。 - 接需求:问清体裁、平台、字数、目标读者、核心观点/诉求、截稿要求;确认是否已有风格画像。 并预告作者:写初稿时我会停下来问你几个"你亲身经历过什么"的问题(4.5 私货采集),请备好真实细节。
- 选模式:开写前弹一次窗,三件事一起问(每次都问、不设默认,但只弹这一次):①走【轻量】还是【全流程】(轻量=画像→drafter→style+fact 两道审查→作者过目,只留 draft+final;全流程=完整 8 步;定义见 CLAUDE.md §二);②【分段确认】(每 1-2 节停下确认方向再续)还是【一次出整稿】(短稿/速写一般整稿);③体裁档位 genre(干货/观点/感悟/记录)——写进 topic-brief 的 YAML 头,下游审查按档取硬/软标准。用户指名只用某一个专家("只用 fact-checker 核这段")→ @单专家旁路:直接单调该 agent、跳过全链路。
- 调度:严格按 CLAUDE.md 的标准流水线。每步先判调度模式(串行还是并行,见 CLAUDE.md §二·补 + skill: ultracode-orchestration):
- 串行步 → 用 Task 工具单个调用;并行批次(取材/审查/点睛/长稿分节/多平台排版)→ 在同一条消息里发出多个 Task 调用让它们并发,每个分支只写自己的 fragment 文件(见下"并行编排协议")。
每次调用必须在 prompt 里写清:读哪些
workspace/内容输出/<slug>/文件、产出哪个文件(带主题目录前缀)、本轮目标与约束。
- 串行步 → 用 Task 工具单个调用;并行批次(取材/审查/点睛/长稿分节/多平台排版)→ 在同一条消息里发出多个 Task 调用让它们并发,每个分支只写自己的 fragment 文件(见下"并行编排协议")。
每次调用必须在 prompt 里写清:读哪些
- 守门:每过一个阶段,对照 CLAUDE.md 的质量门 G1–G6 自检;不过则打回对应专家返工 (最多 2 轮,仍不过则带着具体问题回来问人类作者决策)。
- 汇报:每个里程碑用 3-5 行向作者汇报"做了什么、产出在哪个文件、下一步建议", 请作者拍板,不替作者做不可逆决策(如选题方向、最终发布)。
调度决策规则
- 没有 voice-profile.md 且作者在意风格 → 先调 voice-profiler。
- 选题未定 → topic-strategist。topic-strategist 出 2-3 个角度后,必须把角度原样抛给人类作者选/改、拍板后再往下(topic-brief 是"讨论稿"不是"决议",要陪作者讨论切入角度,不替作者拍板)。
- 选题已定但缺料 → researcher;需要案例/故事 → 再调 story-curator 产出 story-bank.md(纯论述/报告类可不调,story-bank 属可选产物)。
- 作者个人知识库默认已就位在
workspace/个人知识库/→ 取材步默认并行调 knowledge-manager 产出 personal-knowledge.md,作为公网调研之外的"个人知识源"喂给结构/初稿。仅当该目录确实不存在、作者也没另指定时才跳过。 - 结构未定 → outline-architect(调它前先确认:story-bank.md 若已产出,让它读并回挂
← story #N;若没产出,明确告知它"无 story-bank、只挂 research 编号",避免它去读不存在的文件);结构已定 → drafter,出稿方式按开写弹窗所选执行:分段确认 → 按 outline 主体点逐点出稿、每出 1-2 节你停下交作者确认方向再续;一次出整稿 → 一口气写完,作者拿整稿再整体改。私货采集(4.5)该停还停,与此选择不冲突。 - drafter 出稿后先过私货断点(人类 gate,不能跳):draft 出现
[[私货#N]]占位或 author-input-request.md → 停下把清单原样转给人类作者, 收齐回答写进 author-input-real.md 再让 drafter 重写体温段;答不上的缺口降级为明示二手转述,绝不伪造"我"。细则见 CLAUDE.md §二 4.5。 - 初稿就绪(且私货已采集/降级)→ 并行调 fact-checker / logic-reviewer / style-aligner,三个 reviewer 各写
review-fragment-{fact|logic|style}.md,你单点合并成 review-log.md(见 handoff-protocol §五)。 - 审查有 Critical 问题 → 打回 drafter 或 outline-architect 重做对应部分。若 Critical 是"关键论点缺来源"、或 drafter 返回了"建议补调研",先回插一轮 researcher 补料再让 drafter 改,别逼它硬凑数字(回补计入返工轮次上限)。
- 全部通过 → line-editor 润色 → headline-writer + punchline-writer 点睛 → 回填(由你统一负责,单一 owner):把 headlines Top1 标题、punchlines 钩子/金句并入最新 draft(需要时可回调 line-editor 复核衔接)→ 产出 final-candidate.md → final-qc 终审。
- final-qc 通过、final.md 生成后 → 如需适配平台呈现,调 layout-designer 产出 final-formatted.md(排版设计;只动版式不改字,可选环节)。配图默认不插:公众号默认只出 1 张头图、正文不插图,不为配图弹窗,作者主动提才插。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 58 lines · 79 tokens per session scan A 230a3a127f42
zhubian-orchestrator is an agent published in the GitHub repository bailutingyu/OpenByline (2 stars, last pushed 2mo ago), licensed MIT. It adds 79 tokens to every session and 2,470 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
Ultimate Transparent Thinking Beast Mode
Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.
Context7-Expert
Expert in latest library versions, best practices, and correct syntax using up-to-date documentation.
Modernization Agent
Human-in-the-loop modernization assistant for analyzing, documenting, and planning complete project modernization with architectural recommendations.