Skip to content

index.md

Latest commit

 

History

History
145 lines (105 loc) · 5.79 KB

index.md

File metadata and controls

145 lines (105 loc) · 5.79 KB

Multi-Agent Collaboration in Practice: Knowledge Base Sharing and Retrieval

Use case: Team agents, customer support, and multiple business units are all writing documents across different media — the latest version is hard to find, and repeated Q&A is constant. This guide focuses on "unified indexing + layered retrieval + traceable citations," helping you first retrieve the most current and authoritative information, then gradually enrich custom capabilities.

1. What You'll Get

Once it's running, you'll have:

  • Multi-source documents indexed uniformly with automatic tagging
  • Retrieval that surfaces "most relevant + most recent" results first, with source links by default
  • Automatic generation of "summary + citation + recommendation" for each answer, supporting decision-making
  • Multiple agents sharing the same knowledge base, significantly reducing repeated Q&A

2. Copy This Prompt to Claw First

Please help me build a "knowledge base sharing and retrieval" workflow: first pull from three sources — Feishu, Obsidian, and GitHub — sort retrieval results by most recently updated and current status, every answer must include a summary, source link, and expiry status, and automatically generate a "pending revision log" indicating which conflicts require human review.

If you just want to quickly cross-check versions, add: "prioritize returning GitHub docs, since we treat it as the authoritative source."

3. Which Skills You Need

A quick look at what each skill does:

Install with:

clawhub install skill-vetter
clawhub install feishu-doc
clawhub install github
clawhub install summarize
clawhub install obsidian
Skill Purpose
skill-vetter Security scan
feishu-doc Read and extract structured content from Feishu documents
github Sync knowledge from README/PRD/Issues
summarize Compress long documents and extract comparative content
obsidian Local note archiving and long-term management

If you want to connect more private sources, treat the custom retrieval skill in Section 6 as a fallback.

4. What You'll See Once It's Running

[3-Line Summary]
"Customer handoff process" follows Feishu PRD v3 as the authority, last updated 2026-03-20. GitHub docs still says "48 hours" — recommend following the PRD.

[Detailed Citations]
1) Feishu PRD: "Customer Handoff Process v3" 2026-03-20 Wang Wu
2) GitHub docs /handoff.md 2026-02-28 (conflicts with PRD)

[Recommended Action]
"Pending revision log" generated: update "48 hours" to "24 hours" in GitHub docs.

If the output gives conclusions without citations, it means you don't yet have a working skill combination.

5. How to Set It Up Step by Step

Workflow Architecture

flowchart TB
  A["Feishu/Obsidian/GitHub documents"] --> B["Unified extraction + tagging"]
  B --> C["Build index (topic/version/status)"]
  C --> D["Sort retrieval by authoritative source and update time"]
  D --> E["Output summary + citations + conflict notice"]
  E --> F["Generate pending revision log and write back"]
Loading

Configuration Steps

  1. Define the authority source order, e.g., "GitHub docs > Feishu PRD > Obsidian notes."
  2. Standardize metadata fields: topic, status (draft/active/archived), owner, last updated.
  3. In the prompt, require "summary first, then citations, then conflicts/recommendations."
  4. Configure summarize with "max 3-line summary, each with a link, ask again if conflicting."
  5. Use feishu-doc to write back the "pending revision log," recording the source and reason for each update.

6. If No Existing Skill Fits, Have Claw Build One

If your knowledge sources are too private, write a minimal skill:

kb-retrieval/
├── SKILL.md
└── scripts/
    └── search.py

Minimal SKILL.md:

---
name: kb-retrieval
description: Multi-source knowledge retrieval + source citation
---

# KB Retrieval

Invoke when consolidating information from Feishu/GitHub/Obsidian.

The script just needs to "read multiple sources, sort by update time, output structured citations, and generate a pending revision log."

7. Further Optimization

  • Add "owner + update time + status" to each citation so it's easy to tell if it's outdated.
  • Add "explain reasoning when defaulting to authoritative source during conflicts" to the prompt for accountability.
  • Have Claw generate a weekly knowledge update report (e.g., track 5 changes every Friday).

8. Frequently Asked Questions

Q1: Too many retrieval results to read through? A: Show only the top three "authoritative and recent" results first, then offer an "all results" link.

Q2: Conflicting conclusions across sources? A: Add "if conflicting, present the conflict content + suggested resolution" to the prompt.

Q3: Knowledge base growing too large to maintain? A: Set a "status" field; auto-downrank by "draft/active/archived" and auto-archive expired content.

9. Related Reading
⚡