Agent skill

vault

Living knowledge base management. Actions: search (query vault), save (store learning), index (update indices), compile (raw->wiki->rules graduation), init (create vault structure). Follows Karpathy pipeline: ingest->compile->query. Use when: (1) searching accumulated knowledge, (2) saving learnings, (3) compiling raw notes into wiki, (4) initializing a new vault. Triggers: /vault, 'vault search', 'knowledge base', 'save learning'.

Stars 115
Forks 18

Install this agent skill to your Project

npx add-skill https://github.com/alfredolopez80/multi-agent-ralph-loop/tree/main/.claude/skills/vault

SKILL.md

Vault — Living Knowledge Base v3.0

Karpathy-inspired knowledge pipeline: Ingest -> Compile -> Query.

Architecture

The vault is a PRIVATE directory OUTSIDE the public repo:

  • Location: $HOME/Documents/Obsidian/MiVault/ (configurable)
  • Backup: private git repo (user creates their own)
  • Viewer: Obsidian (optional, vault works as plain markdown + git)

This skill teaches HOW to use the vault (framework). It does NOT contain vault data.

Actions

/vault init

Create vault structure:

bash
VAULT_DIR="${VAULT_DIR:-$HOME/Documents/Obsidian/MiVault}"

# 3-layer structure
mkdir -p "$VAULT_DIR"/{global/{raw/articles,raw/papers,raw/images,wiki,output,decisions},projects,_templates}

# Create vault index
echo "# Vault Index" > "$VAULT_DIR/_vault-index.md"
echo "# Project Index" > "$VAULT_DIR/projects/_project-index.md"

Or run: scripts/setup-obsidian-vault.sh

/vault save <classification> <content>

Save a learning to the vault:

Classification Destination Example
GREEN $VAULT_DIR/global/wiki/{category}/ Generic patterns (TypeScript, React)
YELLOW $VAULT_DIR/projects/{project}/wiki/ Project-specific knowledge
RED DISCARDED (never saved) Contains secrets or sensitive info

/vault search <query>

Search vault for relevant knowledge:

  1. Search in: global/wiki/ + projects/{current-project}/
  2. NEVER search in: projects/{other-project}/ (isolation)
  3. Return: matching articles with relevance score

/vault index

Update all _index.md files in the vault:

  1. Scan all .md files in global/wiki/ and projects/*/wiki/
  2. Regenerate _vault-index.md and projects/_project-index.md
  3. Update backlinks between articles

/vault compile

Compile raw notes into wiki articles (Karpathy pipeline):

  1. Read raw/ and lessons/ directories
  2. LLM compiles new articles in wiki/
  3. Update backlinks between articles
  4. Update _index.md
  5. Knowledge Graduation: promote high-confidence learnings to .claude/rules/learned/

Graduation criteria:

  • Confidence >= 0.7
  • Confirmed in >= 3 sessions
  • Format: rule text + source + confidence + last confirmed date

/vault demote <rule>

Revert a graduated rule back to vault if incorrect:

  1. Remove from .claude/rules/learned/{category}.md
  2. Mark source article with status: demoted

Integration Points

Component Integration
session-accumulator.sh PostToolUse hook captures learnings during session
vault-graduation.sh SessionStart hook promotes high-confidence learnings to rules
/exit-review End-of-session GREEN/YELLOW/RED classification
smart-memory-search.sh Adds vault as 5th search source (parallel)
pre-compact-handoff.sh Saves vault context before compaction

Project Isolation

  1. A project ONLY reads: global/wiki/ + projects/{its-own-name}/
  2. NEVER reads projects/{other-project}/ directly
  3. GREEN learnings from any project can be promoted to global/wiki/
  4. SHA-256 deduplication prevents duplicate entries

Anti-Rationalization

Excuse Rebuttal
"The vault is just another notes system" The vault graduates to rules. Notes don't change behavior.
"I'll save it to memory instead" Memory is ephemeral (25KB). Vault is curated knowledge.
"Classification is overhead" 3 seconds to classify saves hours of future searching.
"RED content is useful for context" RED = secrets. Never store. Period.

Expand your agent's capabilities with these related and highly-rated skills.

alfredolopez80/multi-agent-ralph-loop

browser-test

Browser testing using Chrome DevTools MCP and Playwright for visual verification. Start dev server, navigate, screenshot, Lighthouse audit, console errors, network check. Use when: (1) verifying frontend changes, (2) accessibility auditing, (3) performance testing, (4) visual regression. Triggers: /browser-test, 'test in browser', 'visual test', 'lighthouse audit'.

115 18
Explore
alfredolopez80/multi-agent-ralph-loop

crafting-effective-readmes

Use when writing or improving README files. Not all READMEs are the same — provides templates and guidance matched to your audience and project type.

115 18
Explore
alfredolopez80/multi-agent-ralph-loop

stop-slop

A skill for removing AI-generated writing patterns ('slop') from prose. Eliminates telltale signs of AI writing like filler phrases, excessive hedging, overly formal language, and mechanical sentence structures. Use when: writing content that should sound human and natural, editing AI-generated drafts, cleaning up prose for publication, or any content that needs to sound authentic rather than AI-generated. Triggers: 'stop-slop', 'remove AI tells', 'clean up prose', 'make it sound human', 'edit AI writing'.

115 18
Explore
alfredolopez80/multi-agent-ralph-loop

bugs

Bug hunting with Codex CLI Use when: (1) /bugs is invoked, (2) task relates to bugs functionality.

115 18
Explore
alfredolopez80/multi-agent-ralph-loop

adversarial

Multi-Agent Adversarial Analysis System for code security

115 18
Explore
alfredolopez80/multi-agent-ralph-loop

senior-software-engineer

Global skill enforcing senior software engineering best practices

115 18
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results