| name | knowflow-os |
| description | AI-powered knowledge management and content operation system for creators, brands, and content teams.
Manages the full content lifecycle: information collection to knowledge archiving to topic generation to
content production to multi-channel distribution to data review to IP asset accumulation.
Use when: (1) initializing a content or knowledge management project, (2) organizing and classifying
scattered information into reusable knowledge assets, (3) generating content topics from existing
knowledge, (4) producing channel-specific content for Xiaohongshu, WeChat articles, video scripts,
PPT, or Moments, (5) adapting the same content across multiple channels, (6) managing publishing
workflows and post-publish analytics, (7) creating review reports and repurpose recommendations.
Not a traditional CMS. It is an AI-native system where knowledge is grouped by assets, content is
produced by scenarios, and delivery is managed by channel.
|
| metadata | {"author":"wen.yuan"} |
KnowFlow OS
Overview
KnowFlow OS is an AI-native knowledge management and content operation system.
Core loop:
Information Collection -> Knowledge Archiving -> Topic Generation ->
Content Production -> Multi-Channel Distribution -> Data Review -> IP/Brand Asset Accumulation
Core principle:
Knowledge grouped by assets, content produced by scenarios, delivery managed by channel.
Core Capabilities
- Project Initialization: Create the complete KnowFlow OS directory structure with a single command
- Knowledge Import & Classification: Ingest raw materials (text, images, files, links) and classify them into reusable knowledge assets
- Topic Generation: Extract content topics from knowledge assets, scored by relevance and传播 potential
- Channel Content Production: Generate platform-specific content (Xiaohongshu, WeChat, video scripts, PPT)
- Cross-Channel Adaptation: Rewrite the same content for different channels while preserving core messaging
- Publishing & Review: Manage publish workflows, records, and post-publish analytics
Workflow Decision Tree
When the user provides new materials or asks for content creation, follow this decision tree:
User input
├── New material (text/image/file/link)
│ └── Run: import-and-classify workflow
│ -> Save to 00-inbox/
│ -> Extract content
│ -> Classify asset type
│ -> Generate metadata.yml
│ -> Move to 01-knowledge-base/
│
├── "Generate topics" / "Create topic cards"
│ └── Run: knowledge-to-topic workflow
│ -> Select knowledge assets
│ -> Generate topic angles
│ -> Score topics
│ -> Save topic cards to 03-topic-pool/
│
├── "Write content for [channel]" / "Create [platform] post"
│ └── Run: topic-to-content workflow
│ -> Read topic card
│ -> Load related knowledge assets
│ -> Select scenario and channel template
│ -> Generate content draft
│ -> Save to 04-content-factory/
│
├── "Prepare for publish" / "Review before publishing"
│ └── Run: content-to-publish workflow
│ -> Check completeness
│ -> Run review rules
│ -> Move to 05-publish-center/scheduled/
│
├── "Analyze performance" / "Review this content"
│ └── Run: publish-to-review workflow
│ -> Collect publish data
│ -> Analyze performance
│ -> Generate review report to 06-review/
│
└── "Init KnowFlow OS" / "Setup content system"
└── Run: init-project workflow
-> Create full directory structure
-> Create config files (knowflow.yml, AGENTS.md)
-> Create templates and workflows
Quick Start: Project Initialization
When the user asks to initialize a KnowFlow OS project:
-
Create the directory structure:
project-root/
├── 00-inbox/
├── 01-knowledge-base/
│ ├── brand/, products/, courses/, people/, cases/
│ ├── faq/, operations/, training/, research/, media-assets/
├── 02-scenarios/
│ ├── student-content/, course-training/, external-pr/
│ ├── sales-conversion/, community-operation/, teacher-branding/
│ ├── recruitment/, internal-sop/, website-content/
├── 03-topic-pool/
│ ├── pending/, approved/, in-progress/, published/, archived/
├── 04-content-factory/
│ ├── drafts/, pending-review/, approved/, rejected/
├── 05-publish-center/
│ ├── scheduled/, records/
├── 06-review/
│ ├── content-reviews/, channel-reviews/, monthly-reports/
├── 07-ip-assets/
│ ├── brand-identity/, signature-content/, content-series/, reusable-assets/
├── metadata/
├── templates/
├── workflows/
└── skills/
-
Create knowflow.yml from assets/templates/knowflow.yml
-
Create AGENTS.md (see references/agents-rules.md)
-
Copy templates from assets/templates/ to templates/
-
Create workflow configs in workflows/ (see references/workflows-detail.md)
-
Create scenario configs in 02-scenarios/ (see references/project-structure.md)
-
Create README.md
Detailed Workflows
For step-by-step workflow execution, see references/workflows-detail.md.
Workflow 1: import-and-classify
Process: scan_inbox -> detect_type -> extract_text -> summarize -> classify -> generate_metadata -> create_folder -> move -> update_index
Key rules:
- Never delete or overwrite raw files
- All materials must go through 00-inbox/ first
- Generate metadata.yml for every asset
- Create raw.md, summary.md, notes.md for each asset
Workflow 2: knowledge-to-topic
Process: select_assets -> extract_points -> identify_audience -> identify_pain_points -> generate_angles -> score -> save_cards
Key rules:
- Every topic must link to at least one knowledge asset
- Every topic must link to at least one scenario
- Topics should be specific and actionable
Workflow 3: topic-to-content
Process: read_topic -> load_assets -> select_scenario -> select_template -> generate_draft -> generate_titles -> generate_cover -> generate_suggestions -> save
Key rules:
- Different channels require different content versions
- Content must cite which knowledge assets were used
- Include review notes and compliance checks
Workflow 4: content-to-publish
Process: check_completeness -> run_review -> create_version -> move_to_scheduled -> create_checklist
Workflow 5: publish-to-review
Process: collect_data -> analyze -> summarize_feedback -> update_status -> recommend_repurpose -> save_report
Key rules:
- Complete review within 7 days of publishing
- Review must include data and user feedback
- Conclusions must lead to actionable improvements
Asset Categories
Knowledge assets are organized into 10 categories:
| Category | Description |
|---|
| brand | Brand positioning, brand story, visual guidelines |
| products | Product introductions, feature descriptions, pricing |
| courses | Course systems, curricula, teaching methods |
| people | Team introductions, instructor profiles, stories |
| cases | Student/client case studies (anonymized by default) |
| faq | Frequently asked questions, standard answers |
| operations | Operational data, activity records |
| training | Training materials, SOPs, internal knowledge |
| research | Industry reports, competitive analysis, research |
| media-assets | Images, videos, design assets |
Content Channels
Supported channels and their characteristics:
| Channel | Format | Style | Key Requirements |
|---|
| xiaohongshu | 500-1000 words | Casual, emoji-rich | 3-9 images, 3-8 hashtags |
| wechat-article | 1000-3000 words | Structured, in-depth | Summary, cover image |
| moments | 100-300 words | Light, interactive | 1-9 images |
| video-script | Time-based | Conversational | Hook + story + CTA |
| ppt | Slide-based | Visual, point-based | One point per slide |
| brochure | Page-based | Professional | Brand consistent |
| website | SEO-friendly | Formal, concise | Keywords optimized |
| feishu-doc | Document-based | Collaborative | Structured |
Templates
All content templates are in assets/templates/:
topic-card.md — Topic card template
xiaohongshu.md — Xiaohongshu content template
wechat-article.md — WeChat article template
video-script.md — Video script template
ppt-outline.md — PPT outline template
publish-record.md — Publishing record template
review-report.md — Review report template
Examples
A complete yoga studio example is available at assets/examples/yoga-studio/ demonstrating how the same knowledge assets are reused across multiple scenarios.
Core Rules (Summary)
For the complete agent rules, see references/agents-rules.md.
Key principles:
- Raw materials are never deleted or overwritten
- All materials enter through 00-inbox/
- Knowledge assets live in 01-knowledge-base/
- Scenarios in 02-scenarios/, topics in 03-topic-pool/
- Content drafts in 04-content-factory/, publish records in 05-publish-center/records/
- Reviews in 06-review/, IP assets in 07-ip-assets/
- Shared knowledge assets stored once, referenced by scenarios
- Different channels get different content versions
- User cases are anonymized by default
- No exaggerated or unverifiable claims
- Cite knowledge assets used in content generation
- Uncertain classification goes to a "pending confirmation" folder
Resources
references/
assets/