llm-wiki

agent
Security Audit
Warn
Health Warn
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 6 GitHub stars
Code Pass
  • Code scan — Scanned 3 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose

This is a Claude Code plugin and portable agent protocol designed to build, compile, and query topic-specific knowledge bases. It ingests online sources, automates web research via parallel agents, and outputs interconnected Markdown files compatible with tools like Obsidian.

Security Assessment

Overall risk: Medium. The automated code scan of three files found no dangerous patterns, hardcoded secrets, or dangerous permission requests. However, the tool inherently requires significant capabilities to function. Based on the README, it explicitly requests to read and write local files and requires network access to scrape and ingest external URLs. Because the project has low visibility and the exact language/framework is unlisted, a comprehensive security audit is currently impossible. Users should assume it has standard LLM agent capabilities, meaning the host AI model will be instructed to execute file and web operations on your behalf.

Quality Assessment

The project is highly active, with its last push occurring today. Unfortunately, it lacks a software license, which means default copyright laws strictly apply and technically limit legal reuse or modification. Furthermore, it has low community visibility with only 6 GitHub stars. This means the codebase has not been broadly vetted by the open-source community for bugs, edge cases, or security flaws.

Verdict

Use with caution. The core concept and current automation scan are safe, but the lack of licensing and low community validation mean you should inspect the plugin's instructions thoroughly before granting it file and web access.
SUMMARY

A Claude Code plugin for building LLM-compiled knowledge bases. Ingest sources, compile interconnected markdown articles, query, lint, research, and generate outputs — all from Claude Code. Optionally view in Obsidian.

README.md

llm-wiki — nvk's take

A Claude Code plugin for building LLM-compiled knowledge bases. Ingest sources, compile interconnected markdown articles, query, lint, research, and generate outputs — all from Claude Code. Optionally view in Obsidian.

Inspired by Karpathy's LLM wiki concept and his idea file.

Install

Claude Code (native plugin):

/install-plugin github:nvk/llm-wiki

OpenAI Codex / Any LLM Agent (idea file):

# Copy AGENTS.md into your agent's context or project root
cp AGENTS.md ~/your-project/AGENTS.md

The AGENTS.md file contains the complete wiki protocol as a single portable document — works with any LLM agent that can read/write files and search the web.

New to a topic? Two commands. Each topic gets its own wiki — isolated indexes, focused queries, no cross-topic noise:

/wiki init nutrition
/wiki:research "gut microbiome and mental health" --sources 10 --wiki nutrition

The research command launches parallel agents to search the web, ingest the best sources, and compile them into interconnected articles. Drill into gaps with more /wiki:research, ask questions with /wiki:query, drop articles into the inbox anytime.

Quick Start

/wiki init nutrition                              # Create a topic wiki
/wiki:research "gut-brain axis" --wiki nutrition  # 5 parallel agents search, ingest, compile
/wiki:research "probiotics" --deep                # 8 agents, broader angles
/wiki:research "fasting" --retardmax              # 10 agents, max speed, ingest everything
/wiki:query "How does fiber affect mood?"         # Ask the wiki
/wiki:query "compare keto and mediterranean" --deep  # Deep cross-referenced answer
/wiki:ingest https://example.com/article          # Manually ingest a source
/wiki:ingest --inbox                              # Process files dropped in inbox/
/wiki:compile                                     # Compile any unprocessed sources
/wiki:output report --topic gut-brain             # Generate a report
/wiki:output slides --retardmax                   # Ship a rough slide deck NOW
/wiki:assess /path/to/my-app --wiki nutrition     # Gap analysis: repo vs wiki vs market
/wiki:lint --fix                                  # Clean up inconsistencies

Commands

Command Description
/wiki Show wiki status, stats, and list all topic wikis
/wiki init <name> Create a topic wiki at ~/wiki/topics/<name>/
/wiki init <name> --local Create a project-local wiki at .wiki/
/wiki:ingest <source> Ingest a URL, file path, or quoted text
/wiki:ingest --inbox Process all files in the topic wiki's inbox/
/wiki:compile Compile new sources into wiki articles
/wiki:compile --full Recompile everything from scratch
/wiki:query <question> Q&A against the wiki (standard depth)
/wiki:query <question> --quick Fast answer from indexes only
/wiki:query <question> --deep Thorough — reads everything, checks raw + sibling wikis
/wiki:research <topic> 5 parallel agents: academic, technical, applied, news, contrarian
/wiki:research <topic> --deep 8 agents: adds historical, adjacent, data/stats
/wiki:research <topic> --retardmax 10 agents: skip planning, max speed, ingest aggressively
/wiki:search <terms> Find content by keyword or tag
/wiki:lint Run health checks on the wiki
/wiki:lint --fix Auto-fix structural issues
/wiki:lint --deep Web-verify facts and suggest improvements
/wiki:output <type> Generate: summary, report, study-guide, slides, timeline, glossary, comparison
/wiki:output <type> --retardmax Ship it now — rough but comprehensive, iterate later
/wiki:assess <path> Assess a repo against wiki research + market. Gap analysis.
/wiki:assess <path> --retardmax Wide net — adds adjacent fields and failure analysis

All commands accept --wiki <name> to target a specific topic wiki and --local to target the project wiki.

How It Works

Architecture

~/wiki/                                 # Hub — lightweight, no content
├── wikis.json                          # Registry of all topic wikis
├── _index.md                           # Lists topic wikis with stats
├── log.md                              # Global activity log
└── topics/                             # Each topic is an isolated wiki
    ├── nutrition/                      # Example topic wiki
    │   ├── .obsidian/                  # Obsidian vault config
    │   ├── inbox/                      # Drop zone for this topic
    │   ├── raw/                        # Immutable sources
    │   ├── wiki/                       # Compiled articles
    │   │   ├── concepts/
    │   │   ├── topics/
    │   │   └── references/
    │   ├── output/                     # Generated artifacts
    │   ├── _index.md
    │   ├── config.md
    │   └── log.md
    ├── woodworking/                    # Another topic wiki
    └── ...

The hub is just a registry — no content directories, no .obsidian/. All content lives in topic sub-wikis with isolated indexes and articles. Queries stay focused. The multi-wiki peek finds overlap across topics when relevant.

The Flow

  1. Research a topic — parallel agents search the web, ingest sources, and compile articles in one command
  2. Ingest additional sources — URLs, files, text, tweets (via Grok MCP), or bulk via inbox
  3. Compile raw sources into synthesized wiki articles with cross-references and confidence scores
  4. Query the wiki — quick (indexes), standard (articles), or deep (everything)
  5. Assess a repo against the wiki — gap analysis: what aligns, what's missing, what the market offers
  6. Lint for consistency — broken links, missing indexes, orphan articles
  7. Output artifacts — summaries, reports, slides — filed back into the wiki

Key Design

  • One topic, one wiki — each research area gets its own sub-wiki with isolated indexes. No cross-topic noise.
  • Parallel research agents — 5 standard, 8 deep, 10 retardmax. Each agent searches from a different angle.
  • _index.md navigation — every directory has an index. Claude reads indexes first, never scans blindly.
  • Articles are synthesized, not copied — they explain, contextualize, cross-reference.
  • Raw is immutable — once ingested, sources are never modified.
  • Multi-wiki aware — queries peek at sibling wiki indexes for overlap.
  • Dual-linking — both [[wikilinks]] (Obsidian) and standard markdown links on every cross-reference. Works everywhere.
  • Confidence scoring — articles rated high/medium/low based on source quality and corroboration.
  • Structural guardian — auto-checks wiki integrity after operations, fixes trivial issues silently.
  • Activity loglog.md tracks every operation, append-only, grep-friendly.
  • Zero dependencies — runs entirely on Claude Code built-in tools.

Research Modes

Mode Flag Agents Style
Standard (default) 5 Academic, technical, applied, news, contrarian
Deep --deep 8 Adds historical, adjacent fields, data/stats
Retardmax --retardmax 10 Adds rabbit-hole agents. Skip planning, cast widest net, ingest aggressively, compile fast. Lint later.

Retardmax mode is inspired by Elisha Long's retardmaxxing philosophy — act first, think later. The antidote to analysis paralysis. Works for both /wiki:research and /wiki:output.

Linking: Works Everywhere

Every cross-reference in the wiki uses dual-link format:

[[gut-brain-axis|Gut-Brain Axis]] ([Gut-Brain Axis](../concepts/gut-brain-axis.md))

The wiki is not locked into any tool:

  • Obsidian reads the [[wikilink]] — graph view, backlinks panel, quick-open
  • Claude Code follows the standard (relative/path.md) link
  • GitHub/any markdown viewer renders the standard link as clickable
  • No viewer at all — plain markdown, readable in any text editor

Obsidian Integration

Each topic wiki has its own .obsidian/ config and can be opened as an independent vault:

open ~/wiki/topics/nutrition/     # Open in Obsidian — focused graph for one topic

The hub (~/wiki/) has no .obsidian/ to avoid nested vault confusion. If you want a cross-topic view, open ~/wiki/ manually and let Obsidian create its own config.

What works out of the box:

  • .obsidian/ config created on init with sane defaults
  • [[wikilinks]] power the graph view
  • aliases in frontmatter enable search by alternate names
  • tags in frontmatter are natively read
  • inbox/ works as a drop zone in both Obsidian and the CLI

Claude Code is the compiler. Obsidian is an optional viewer.

Query Depths

Depth Flag What it does
Quick --quick Reads indexes only. Fastest. For simple lookups.
Standard (default) Reads relevant articles + full-text search. For most questions.
Deep --deep Reads everything, searches raw sources, peeks sibling wikis. For complex questions.

Credits

License

MIT

Reviews (0)

No results found