kb-query

KB Query — Knowledge Search + Synthesis

Search the Knowledge Base, synthesize answers from multiple documents, cite sources with wikilinks, and optionally accumulate new insights as vault documents.

Usage

/kb-query "list all papers related to ML" (또는 "k8s 관련 논문 전체 목록")
/kb-query "my-webapp deployment status" (또는 "bilingua 배포 상태")
/kb-query "documents referencing MyProject" (또는 "ChaosEater를 참조하는 문서")
/kb-query "summary of work in last 2 weeks" (또는 "최근 2주 작업 요약")

Route Selection

Choose the optimal search path by question type. Do not always start from the top-level index — pick the fastest route.

Inbox Structure Notes

• Treat Inbox/**/README.md as structural guidance notes for folder usage.
• Exclude or de-prioritize these README files during routine knowledge queries.
• Use them when the question is specifically about inbox structure, folder naming, or note organization.

Route A: Direct Folder Access

When: Target is clearly in one PARA category.

"my-webapp deployment status" → read 1. Projects/my-webapp/ directly
"paper list" → read 2. Areas/Paper/_index.md then scan Paper/

Route B: Tag Cross-Collection

When: Need to gather documents across folders by topic.

"everything about my-project" → search for #proj/my-project OR #topic/kubernetes across vault
"ML optimization related" → search #topic/ml-optimization

Route C: Backlink Traversal

When: Need to find what references a specific document or entity.

"where is MyProject used" → search for [[MyProject]] and "MyProject"
"documents referencing this paper" → search for [[Paper Title]]

With obsidian CLI:

obsidian backlinks file="DocumentName"

Without CLI: Grep for [[DocumentName]] across vault.

Route D: Index + Log Browse

When: Broad overview or temporal queries.

"최근 뭐 했지" → read 0. Common/log.md (last 20 entries)
"전체 현황" → read 0. Common/index.md
"Projects 개요" → read 1. Projects/_index.md

Route E: Full-Text Search

When: Specific term or concept lookup.

"Prometheus metric setup" → Grep "Prometheus" across vault
"gradient descent" → Grep "gradient.descent" across vault

Combined Routes

Complex questions may combine routes:

"key techniques in papers referenced by my-project"
  Route A: read project docs for paper references
  Route A: read each referenced paper
  Synthesize: compare techniques

Execution Flow

1. PARSE

• Extract: subject, scope, temporal range, expected output type
• Select route(s): A, B, C, D, E, or combination
• Estimate token budget (stay under 10K tokens of source reading)

2. SEARCH

• Start with _index.md for orientation (10-20 lines, cheap)
• Read full documents only when needed for the answer
• Use Grep for targeted term search
• Exclude or de-prioritize Inbox/**/README.md unless the query is about structure or workflow
• Expand with backlinks if initial results are insufficient

3. SYNTHESIZE

• Combine information from multiple sources
• Always cite with [[wikilinks]]
• Format by output type:
  • Lists: include file locations
  • Analysis: include evidence and reasoning
  • Timelines: chronological order
  • Comparisons: structured table

4. ACCUMULATE (optional)

Decide whether the result is worth saving:

Result Type	Action
Simple factual answer	Respond only
New insight or synthesis	Offer to save as document
Cross-document comparison	Suggest saving
List/catalog	Suggest if useful for reuse

If saving:

• Choose PARA location
• Create with proper frontmatter and tags
• Add wikilinks to source documents
• Update _index.md and log.md
• Backlink from source documents

5. RESPOND

Format:

## Answer
{Direct answer}

### Sources
- [[Doc 1]] — relevant detail
- [[Doc 2]] — relevant detail

### Related
- [[Other Doc]] — might be useful

Obsidian CLI Integration

When obsidian is available (Obsidian 1.12+), prefer it for these operations:

# Route B: tag-based collection
obsidian tags sort=count counts          # tag inventory
obsidian search query="#proj/my-project"  # tag search

# Route C: backlink traversal
obsidian backlinks file="MyProject"      # precise backlinks

# Route E: full-text search
obsidian search query="Prometheus" limit=10   # native vault search

# Reading content
obsidian read file="Note Name"           # read by wikilink name
obsidian read path="1. Projects/_index.md"  # read by path

Detection: Check which obsidian or try direct call. If unavailable, fall back to Grep/Glob/Read tools.

Complementary with kepano/obsidian-skills: The obsidian-cli SKILL.md from kepano provides the full CLI reference. This skill focuses on when and why to use each command for knowledge queries.

Token Efficiency

• Read _index.md first (~10-20 lines) before full documents
• Use CLI search or Grep to find specific sections, not read entire files
• For large documents, use offset + limit to read relevant portions
• Prioritize: index → frontmatter → target section → full document