Skill: knowledge-base

What I do

I teach agents how to access the three knowledge systems available in this setup: the memory graph (MCP), the Obsidian vault via RAG, and direct KB doc navigation. I prevent re-discovering what's already documented.

When to use me

• Before starting any investigation — check what's already known
• When a skill's ## KB Reference points to a KB doc you need to read
• When searching for past decisions, patterns, or solutions
• When you need context about a codebase, agent, skill, or workflow

The three knowledge systems

Patterns & examples

Search memory graph (fastest — check first):

mcp_memory_search_nodes({ query: "describe the problem or topic" })
mcp_memory_open_nodes({ names: ["KnownEntityName"] })

Query vault via RAG (semantic search across all docs):

mcp_vault-rag_query_vault({
  vault: "baphled",
  question: "what is the pattern for X?",
  top_k: 5
})

Read KB doc directly (when you know the path):

~/vaults/baphled/3. Resources/Knowledge Base/Skills/{Category}/{Name}.md
~/vaults/baphled/3. Resources/Tech/OpenCode/
~/vaults/baphled/3. Resources/Knowledge Base/Agents/{Name}.md

Lookup order

1. Memory graph — search_nodes for the topic
2. Vault-RAG — query_vault if memory has nothing
3. Direct KB read — if you know the exact doc path
4. Codebase investigation — only if none of the above answers it

Anti-patterns to avoid

• ❌ Investigating the codebase before checking memory/vault
• ❌ Asking the user for context that's already in the KB
• ❌ Ignoring ## KB Reference sections in skills — they point to deeper coverage
• ❌ Storing to memory without searching first (creates duplicates)

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Core-Universal/Knowledge Base.md

Related skills

• memory-keeper — Capturing and retrieving from the memory graph
• obsidian-structure — PARA structure for navigating the vault
• investigation — Systematic codebase investigation when KB has no answer