| name | llmwiki-second-brain |
| description | Scaffold and configure an LLMWiki-style AI-maintained second brain. Use when the user wants to create, install, clone, set up, initialize, or automate a personal Obsidian/Claude/Codex knowledge vault; convert the LLMWiki playbook into an agent-run workflow; create a second brain from this GitHub repo; or avoid manually following the README steps. |
LLMWiki Second Brain
This skill helps Codex create an LLMWiki-style second brain vault from the bundled starter templates.
Use the script first whenever possible. Then guide the user through the remaining app connections.
Quick Workflow
- Choose a vault path.
- If the user gives a path, use it.
- If not, use
~/Documents/LLMWiki.
- Run
scripts/create_llmwiki_vault.py <vault-path> from this skill directory.
- Tell the user what was created and what was skipped.
- Tell the user to open the created folder as an Obsidian vault.
- Tell the user to connect Claude Desktop to that folder with filesystem access.
- Tell the user to open the same folder in Codex so
AGENTS.md is active.
- Give the user one test prompt:
Read CLAUDE.md and wiki/index.md, then confirm you understand the LLMWiki operating rules.
Script
Run:
python3 <skill-dir>/scripts/create_llmwiki_vault.py ~/Documents/LLMWiki
Useful flags:
--include-examples copies the fictional example raw/wiki pages.
--force overwrites existing starter files.
--dry-run prints planned operations without writing.
The script creates:
CLAUDE.md
AGENTS.md
raw/assets/
wiki/index.md
wiki/log.md
wiki/entities/
wiki/concepts/
wiki/sources/
wiki/syntheses/
.claude/commands/ingest.md
.claude/commands/query.md
.claude/commands/lint.md
After Scaffolding
Explain the operating loop:
capture -> ingest -> query -> synthesize -> lint
Tell the user:
- Save source material into
raw/.
- Ask Claude or Codex to ingest important raw files.
- Ask questions from the wiki, not from model memory.
- File useful answers into
wiki/syntheses/.
- Run
/lint regularly.
Claude Desktop Connection
If the user asks for MCP setup, give the filesystem server config and replace the path with their vault path:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/absolute/path/to/LLMWiki"
]
}
}
}
On Windows, use cmd /c npx and a Windows path.
Security rule: grant access only to the vault folder, not the whole home directory.
GitHub Repo Flow
If the user wants to distribute this:
- Keep private vault content out of Git.
- Publish only the playbook, templates, examples, and this skill.
- Tell users they can clone the repo, copy
skills/llmwiki-second-brain into their Codex skills directory, then ask Codex:
Use the LLMWiki Second Brain skill to create my vault at ~/Documents/LLMWiki.
Guardrails
- Do not overwrite an existing vault unless the user explicitly asks for
--force.
- Do not ingest private content while scaffolding.
- Do not put user secrets, API keys, or personal raw notes into a public repo.
- Keep all internal wiki references as Obsidian wikilinks.
- Keep slugs lowercase and hyphen-separated.