| name | knowledge-organizer |
| license | MIT |
| description | An OpenClaw- and Codex-compatible knowledge organization skill for importing articles, organizing notes, syncing to Feishu and Tencent IMA, applying tags, archiving content, generating summaries, and suggesting related notes. |
| env | ["OPENCLAW_KB_ROOT","FEISHU_APP_ID","FEISHU_APP_SECRET","FEISHU_ACCESS_TOKEN","FEISHU_KB_ID","FEISHU_FOLDER_ID","FEISHU_WIKI_NODE","IMA_OPENAPI_CLIENTID","IMA_OPENAPI_APIKEY"] |
Knowledge Organizer
This skill turns article links, drafts, and notes into structured Markdown with duplicate checks, tags, summaries, related-note suggestions, image downloads, and optional sync targets for Obsidian, Feishu, and Tencent IMA.
Use Cases
- Store content in a knowledge base
- Organize articles
- Apply tags
- Archive notes
- Generate summaries
- Suggest related notes
Capabilities
- Generate Obsidian-ready notes with YAML frontmatter, wikilinks, embeds, and block IDs
- If
draft.images is present, download images into assets/ and keep relative references in the note body; common fields like src, data_src, data-original, data-lazy-src, srcset, url, image_url, and original are supported
- Run duplicate detection before writing, covering URL, hash, alias, and similarity checks
- Treat duplicate hits as normal control flow; the CLI returns a structured decision result
- Recommend directly linkable related notes
- Validate tags against the knowledge-base tag contract
- Sync to Feishu through the official OpenClaw
openclaw-lark plugin
- Sync to Tencent IMA through the direct import_doc OpenAPI flow
- Orchestrate
destination=obsidian|feishu|ima with mode=once|sync
Usage
- For Obsidian, set
destination=obsidian, provide vault_root, and let the runtime write directly to local markdown files.
- For Feishu, set
destination=feishu and make sure the OpenClaw openclaw-lark plugin is available for feishu-create-doc and feishu-update-doc.
- For Tencent IMA, set
destination=ima and configure IMA_OPENAPI_CLIENTID and IMA_OPENAPI_APIKEY before syncing.
- Use
mode=once for a single import run, or mode=sync to skip items whose content hash has not changed.
- The shared sync orchestrator lives in
scripts/knowledge_sync.py, and can be used with --markdown-path, --folder-path, or --link.
Usage Examples
- Obsidian:
python3 -m scripts.knowledge_sync --destination obsidian --mode once --state .sync-state.json --vault-root /path/to/vault --markdown-path draft.md
- Feishu:
python3 -m scripts.knowledge_sync --destination feishu --mode once --state .sync-state.json --markdown-path draft.md
- Tencent IMA:
python3 -m scripts.knowledge_sync --destination ima --mode sync --state .sync-state.json --folder-path drafts/
- Disable one or more destinations: add
--disable obsidian,feishu to skip the selected targets when invoking the sync orchestrator.
Workflow
- Get content: use a browser for public-account links, prefer
xiaohongshu-mcp for Xiaohongshu links, use web_fetch for other web pages, and process user-provided content directly
- Normalize the source: use
scripts/import_sources.py, scripts/import_normalizer.py, and scripts/import_models.py to build a shared ImportDraft
- Check duplicates before final write: prefer URL + title + similarity checks, and treat duplicate hits as normal control flow
- Choose a destination:
scripts/knowledge_sync.py dispatches to obsidian, feishu, or ima
- Render or import:
scripts/obsidian_note.py generates the content and destination path, while scripts/feishu_kb.py and scripts/ima_kb.py build the sync payloads
- Write or sync: runtime writes directly to
destination_path for Obsidian, or stores SyncStateRecord for Feishu/IMA
For WeChat public-account imports, read references/wechat-import.md before doing browser extraction, image handling, or final write.
Execution Rules
- After reading this skill, follow its workflow before improvising your own path
- Do not skip duplicate checks on import tasks
- If a browser/evaluate call fails twice because of parameter misuse, stop blind retries and re-read the workflow / reference
- Prefer reusing bundled scripts over hand-writing a parallel pipeline when the scripts already cover the task
- For article imports, separate the pipeline into: fetch/normalize → duplicate-check → render → write, instead of mixing all steps together
- Before running script-based import flows, prefer checking
scripts/check_runtime.py to confirm Python and knowledge-base paths are available
- For Feishu import, prefer the official OpenClaw
openclaw-lark plugin (feishu-create-doc / feishu-update-doc) and use Lark Markdown with <image url="..."/> / <file url="..." name="..."/>
- For IMA import, use the direct OpenAPI
import_doc flow with IMA_OPENAPI_CLIENTID and IMA_OPENAPI_APIKEY
Contract
- Input: structured draft, title aliases, source metadata, summary, bullets, excerpts, images, related notes, and vault root
- Output:
RenderedNote(content, destination_path) for Obsidian, or a SyncStateRecord-backed import result for Feishu/IMA
- Frontmatter must include
title, aliases, tags, source_type, source_url, published, created, updated, importance, status, and canonical_hash
- Before writing tags, require at least 1 domain tag and 1 type tag, with a total of 5-10 tags
- Vault root must be a non-empty absolute path
- Vault root should come from
OPENCLAW_KB_ROOT when available
- This contract covers frontmatter / wikilink / embed / block id rules
- Feishu payloads use
title + markdown + placement fields (folder_token / wiki_node / wiki_space)
- IMA payloads use
content_format=1, content, and optional folder_id
WeChat Notes
- WeChat article links (
mp.weixin.qq.com) are a special case: default to browser, not web_fetch
- Prefer the container order documented in
references/wechat-import.md when extracting正文
- For image extraction, prefer
data-src, then src, then data-original / data_src / original
- Preserve or add
from=appmsg on WeChat image URLs when needed
- Normalize image fields before conversion when possible (treat resolved
data-src as the final src)
- Strip common tail noise (scan prompts, 授权提示, 阅读原文引导) before final render
browser act with evaluation must include fn; missing fn is a caller error, not a reason to improvise blindly
Xiaohongshu Notes
- Xiaohongshu links are a special case: default to
xiaohongshu-mcp, not generic web_fetch
- Prefer checking MCP status first when the workflow depends on local login/session
- Prefer
detail for complete note content; use search → detail only when direct note identifiers are missing
- Treat Xiaohongshu as a structured content source, not just a webpage snapshot
- Preserve source metadata such as author, publish time, tags, images, and engagement when available
draft.images Example
images:
- path: /absolute/path/to/local.png
alt: Local image
- src: https://example.com/cover.png
alt: Remote image
- srcset: https://example.com/cover-1x.png 1x, https://example.com/cover-2x.png 2x
alt: Responsive image
path is for local files. src / data_src / data-original / data-lazy-src / original etc. are used for remote images; srcset prefers the highest-value candidate.
Compatibility
- OpenClaw 兼容
- Codex 兼容
- Obsidian vault 工作流
Project Links
Output Template
✅ Stored in knowledge base
📁 Location: knowledge-base/xxx.md
🏷️ Tags: tag1, tag2, tag3
📋 Summary: one-sentence summary
⭐ Importance: core
🔗 Related notes: xxx, yyy
