← 返回
内容创作 中文

Doc Orchestrator

Orchestrate multi-chapter document generation using sub-agents. Use when producing long structured documents (PRDs, technical specs, research reports, design...
使用子代理编排多章节文档生成,适用于长结构化文档(如PRD、技术规格、研究报告、设计等)。
nicolayao
内容创作 clawhub v2.0.0 1 版本 99863.2 Key: 无需
★ 0
Stars
📥 730
下载
💾 20
安装
1
版本
#latest

概述

Doc Orchestrator

Generate long, multi-chapter documents by coordinating sub-agents with a contract-first, serial-then-parallel strategy and persistent orchestration state.

When This Applies

  • Document has 3+ chapters/sections with cross-references
  • Total expected output > 500 lines
  • Multiple sections share definitions (names, enums, constants, values)

Core Principles

  1. Contract First — Define all shared definitions before delegating
  2. File Isolation — Each sub-agent writes to its own file; main agent merges
  3. State Persistence — Write orchestration state to JSON so context compaction can't break the workflow

Workflow

Phase 1 — Analyze Dependencies

Build a dependency graph. For each pair ask: "Does chapter B need chapter A's output?"

Classify:

  • Contract — defines shared values (main agent writes directly)
  • Serial — has upstream dependency (spawn after dependency completes)
  • Parallel — no unresolved dependencies (run concurrently)

Optimize: if multiple chapters depend only on the contract, they can run in parallel.

Phase 2 — Initialize State File

Create {task-dir}/{TASK-ID}-orchestration.json:

{
  "document": "Document Title",
  "contract_file": "TASK-XXX-contract.md",
  "final_file": "TASK-XXX-final.md",
  "chapters": {
    "ch1": {"title": "Overview", "status": "done", "file": "TASK-XXX-contract.md", "deps": []},
    "ch2": {"title": "Characters", "status": "pending", "file": "TASK-XXX-ch2.md", "deps": ["ch1"]},
    "ch3": {"title": "Abilities", "status": "pending", "file": "TASK-XXX-ch3.md", "deps": ["ch1"]},
    "ch4": {"title": "Factions", "status": "pending", "file": "TASK-XXX-ch4.md", "deps": ["ch1", "ch2"]}
  }
}

Status values: pending | running | done | failed

Update this file after every state change. After context compaction, read this file to restore full state.

Phase 3 — Write the Contract

Main agent writes all contract chapters directly. Include a Global Conventions table:

## Global Conventions
| Item        | Value       | Referenced by  |
|-------------|-------------|----------------|
| Score range | 1-5 integer | ch5 API, ch6   |

This is the single source of truth.

Phase 4 — Execute (Serial + Parallel)

For each chapter whose deps are all done:

  1. Update state: "status": "running"
  2. Spawn sub-agent with the prompt template (see below)
  3. On completion: update state to "done"; check what new chapters are unblocked
  4. On failure: update state to "failed"; decide retry or main-agent fallback

Spawn all unblocked chapters in parallel. Wait for serial dependencies.

After context compaction: read the orchestration JSON to recover state, then continue.

Phase 5 — Merge & Validate

Concatenate files in chapter order. During merge:

  1. Strip duplicate document titles — Sub-agents often repeat the top-level # Document Title. Remove all occurrences except the one in the contract file:

```bash

# Remove duplicate H1 titles during merge (keep only from contract)

cat contract.md > final.md

for f in ch2.md ch3.md ... ; do

sed '/^# Document Title$/d' "$f" >> final.md

done

```

  1. Run consistency checks:

```bash

grep -n "conflicting_value" final.md

grep -o "'[a-z_]*'" final.md | sort | uniq

```

  1. Fix any issues before delivering.

Sub-Agent Prompt Template

## Task: Write [Chapter Title] for [Document Name]

**Read first:** `path/to/contract.md`
Pay attention to Global Conventions table (section X.X).

**Write to:** `path/to/output-chN.md` (new file, do NOT modify other files)

**IMPORTANT formatting rules:**
- Do NOT include the document title (# Document Title) — it belongs only in the contract
- Start your file directly with the chapter heading (## Chapter N: Title)
- Do NOT repeat definitions from the contract; reference them

### Content requirements:
[chapter-specific requirements]

### Constraints (must match contract):
- [constraint 1]
- [constraint 2]

Anti-Patterns

Don'tDo Instead
------------------
Rely on context memory for orchestration statePersist state to JSON file
Let sub-agents write to the same fileEach writes own file; main agent merges
Skip formatting rules in promptExplicitly say "no document title, start with ## chapter heading"
Assume sub-agents won't hit content filtersHave fallback: main agent writes sensitive chapters directly
Skip consistency check after mergeAlways grep for known conflict patterns
Use bigger model to brute-force long outputSmaller model + smaller task > bigger model + huge task
Poll sub-agents in a loopUse push-based completion (auto-announce)

Decision Flowchart

Document request received
  |
  +-- < 500 lines expected? --> Write directly (no orchestration)
  |
  +-- 500-1500 lines, no cross-refs? --> Simple parallel (each chapter = own file)
  |
  +-- > 500 lines WITH cross-references?
       |
       1. Analyze dependency graph
       2. Create orchestration state JSON
       3. Main agent writes contract chapters
       4. Serial chain for dependent chapters
       5. Parallel burst for independent chapters
       6. Merge + strip duplicate titles + validate
       7. Deliver

Lessons from Real Usage

Test 1: PRD (9 chapters, technical)

MetricNaive Parallel (v1)Contract-First (v2)
------------------------------------------------
Output1,405 lines3,055 lines
ConsistencyScore 1-5 vs 1-10 conflictZero conflicts
File integrityChapters overwrittenAll preserved
Rework4 chapters rewrittenNone

Test 2: Worldbuilding Bible (7 chapters, creative writing)

MetricResult
----------------
Output2,170 lines
ConsistencyZero conflicts (names, factions, abilities all matched)
Issues hitContent filter blocked ch5 twice; ch6 output wrong chapter content
RecoveryMain agent wrote ch5 directly; ch6 retried successfully
Duplicate titles4 of 6 sub-agent files repeated doc title (fixed in merge)

Key Takeaways

  • Content filters: Sensitive topics (war, conflict) may trigger model safety filters in sub-agents. Fallback: main agent writes those chapters, or rephrase with softer language
  • Wrong output: Sub-agents occasionally output content for the wrong chapter. Always verify file content, not just file existence
  • Title duplication: Sub-agents copy the document title from the contract file. Prompt template now explicitly forbids this; merge step strips duplicates as safety net

版本历史

共 1 个版本

  • v2.0.0 当前
    2026-03-30 07:49 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

content-creation

YouTube

byungkyu
使用托管OAuth集成YouTube Data API,支持搜索视频、管理播放列表、获取频道数据及评论互动,适用于用户需要时使用此技能。
★ 142 📥 41,077
content-creation

Baidu Wenku AIPPT

ide-rea
使用百度文库 AI 智能生成 PPT,自动根据内容选择模板。
★ 66 📥 46,208
content-creation

AdMapix

fly0pants
广告情报与应用数据分析助手,支持搜索广告素材、分析应用排名、下载量、收入及市场洞察,用于广告素材和竞品分析。
★ 295 📥 136,494