LLM Wiki

Karpathy's LLM Wiki

Build and maintain a persistent, compounding knowledge base as interlinked markdown files. Based on Andrej Karpathy's LLM Wiki pattern.

Unlike traditional RAG (which rediscovers knowledge from scratch per query), the wiki compiles knowledge once and keeps it current. Cross-references are already there. Contradictions have already been flagged. Synthesis reflects everything ingested.

Division of labor: The human curates sources and directs analysis. The agent summarizes, cross-references, files, and maintains consistency.

When This Skill Activates

Use this skill when the user:

• Wants to research a topic (auto-create wiki if none exists, then proceed with research)
• Asks to create, build, start, or structure a new wiki, note system, or knowledge base
• Asks to ingest, add, or process a source into their wiki
• Asks a question and an existing wiki is present
• Asks to lint, audit, or health-check their wiki
• References their wiki, knowledge base, or "notes" in a research context

The wiki is research infrastructure, not the end goal. Users want research results; the wiki is the mechanism that makes research compounding and persistent. When a user says "help me research X" and no wiki exists, create one and proceed with the research.

Wiki Location

The wiki is just a directory of markdown files — open it in Obsidian, VS Code, or any editor. No database, no special tooling required.

Path Resolution

1. User specifies a path — use it directly.
2. No path given — analyze the context to suggest paths:
   • What is the user researching? What kind of directory name fits this topic?
   • What does the CWD directory structure look like? Are there existing directories that suggest a naming convention?
   • Suggest 2-3 contextually appropriate options, then let the user confirm or provide their own.
3. Never scan ~/ — do not search the home directory for existing wikis.
4. Never assume a default path — always resolve explicitly.

User provides path → use it
    ↓ (no path)
Analyze context (topic, CWD structure, existing dirs)
    ↓
Suggest 2-3 fitting options → user confirms or gives own path
    ↓
Use confirmed path for all subsequent operations

Architecture: Three Layers

wiki/
├── SCHEMA.md           # Conventions, structure rules, domain config
├── index.md            # Global catalog: topics + cross-topic pages
├── log.md              # Chronological action log (append-only, rotated at 500 entries)
├── raw/                # Layer 1: Immutable source material
│   ├── articles/       # Web articles, clippings
│   ├── papers/         # PDFs, arxiv papers
│   ├── transcripts/    # Meeting notes, interviews
│   └── assets/         # Images, diagrams referenced by sources
├── topics/             # Layer 2: Per-research-topic grouping
│   ├── {topic-slug}/
│   │   ├── concepts/   # Concepts specific to this topic
│   │   ├── entities/   # Entities specific to this topic
│   │   ├── comparisons/ # Comparisons specific to this topic
│   │   └── index.md    # Topic-level catalog
│   └── ...
├── concepts/           # Layer 2: Cross-topic concept pages
├── entities/           # Layer 2: Cross-topic entity pages
├── comparisons/        # Layer 2: Side-by-side analyses
└── queries/            # Layer 2: Filed query results worth keeping

Layer 1 — Raw Sources: Immutable. The agent reads but never modifies these. Layer 2 — The Wiki: Agent-owned markdown files. Created, updated, and cross-referenced by the agent. Organized by research topic. Layer 3 — The Schema: SCHEMA.md defines structure, conventions, and tag taxonomy.

Topic Grouping Rules

Each research topic gets its own subdirectory under topics/. When ingesting:

• Single-topic page (only relevant to one research area) → topics/{topic}/concepts/, topics/{topic}/entities/, or topics/{topic}/comparisons/
• Cross-topic page (relevant to 2+ research areas) → global concepts/, entities/, or comparisons/
• Topic pages use [[wikilink]] to reference global pages and vice versa

判断标准: When a new entity/concept appears in a source, check if it's already referenced in another topic. If yes → global. If no → place in the current topic.

Two-Level Index

• Global index.md — Lists all research topics (linking to their index) + cross-topic pages
• Topic index.md — Lists all pages within that specific topic

Note Discipline (Zettelkasten)

Wiki pages follow Zettelkasten（卡片盒筆記法）discipline — refer to the zettelkasten skill for the full method. Mapping:

• Concept and entity pages are cards — one page = one distinct idea/entity. Split by idea count, not just line count.
• Global/topic index.md are structure notes — navigation, atomicity does not apply. The structure-note role is carried entirely by the two-level index: don't create separate structure-note files in a wiki. Exception: a split remnant page ("summary + links", zettelkasten split procedure step 4) is a small concept-type page, not a structure-note violation.
• Comparison and query pages are synthesis notes — they exist to combine cards, so atomicity does not apply, but linking rules do. Ideas born inside a synthesis page that stand on their own must be promoted to concept pages (see Page Thresholds) — synthesis pages must not become a shelter from atomicity.
• raw/ plays the literature-note role — sources stay immutable; wiki pages restate claims in their own words and cite provenance.

Every page create or update runs this checklist before writing. Authoritative version with full split criteria and worked cases: the zettelkasten skill (references/02-split-and-checklist.md); the core below is the fallback when that skill is unavailable:

1. Atomicity — is this addition a single idea (entity pages: a single entity)?
2. Split decision — split when any criterion hits: the title needs "and" to cover the content; a passage could be cited independently from another context; the page answers more than one question; parts update at different rates; inbound links can't say which point they target. Don't split when pieces would lose self-containment, are only ever cited together, or are mere rephrasings or examples of the same idea. Length (~200 lines) is a warning to run these criteria, not a reason to split.
3. Duplicate & conflict check — search first. Existing idea → update that page. Conflicting information → don't merge-overwrite; follow the Update Policy (note both positions, mark contradictions/contested, link both ways).
4. Links — link terms in place where the text mentions them (1-2 real outbound links); relations the text doesn't naturally mention go in a labeled end-section; check whether linked pages need a link back. Backlink edits don't recurse: fixing a backlink doesn't trigger a checklist run on that page.
5. Structure/index — add or refresh the page's row in the topic/global index.
6. Self-contained — the page stands alone without the source context.
7. Own words — restate, don't paste; cite provenance.

This is a per-write discipline, not a lint-time cleanup.

Deference: this discipline governs llm-wiki wikis. When working in a notes repository that is not an llm-wiki (no SCHEMA.md/index.md), follow that repository's own conventions — mirroring the zettelkasten skill's 沿用現場 rule.

Resuming an Existing Wiki (CRITICAL — do this every session)

When the user has an existing wiki, always orient yourself before doing anything:

① Read SCHEMA.md — understand the domain, conventions, and tag taxonomy. ② Read index.md — learn what pages exist and their summaries. ③ Scan recent log.md — read the last 20-30 entries to understand recent activity. ④ Quick consistency probes (cheap, catches drift before it compounds): compare the actual page count against the index header's "Total pages"; and spot-check the wiki's SCHEMA.md for rules superseded by the current skill canon (e.g., "split at 200 lines" as a criterion, "minimum 2 links", missing comparison/query exemptions). If either probe fails, suggest running lint or updating SCHEMA.md before writing.

# Orientation reads at session start (replace <wiki> with the resolved path)
read_file "<wiki>/SCHEMA.md"
read_file "<wiki>/index.md"
read_file "<wiki>/log.md" offset=<last 30 lines>

Only after orientation should you ingest, query, or lint. This prevents:

• Creating duplicate pages for entities that already exist
• Missing cross-references to existing content
• Contradicting the schema's conventions
• Repeating work already logged

For large wikis (100+ pages), also run a quick search_files for the topic at hand before creating anything new.

Initializing a New Wiki

When the user asks to create or start a wiki:

1. Resolve the wiki path (see Path Resolution above)
2. Create the directory structure above
3. Ask the user what domain the wiki covers — be specific
4. Ask what research topics they plan to work on (creates topics/ subdirectories)
5. Write SCHEMA.md customized to the domain (see template below)
6. Write initial index.md with topic listings
7. Write initial log.md with creation entry
8. Confirm the wiki is ready and suggest first sources to ingest

SCHEMA.md Template

Adapt to the user's domain. The schema constrains agent behavior and ensures consistency:

# Wiki Schema

## Domain
[What this wiki covers — e.g., "AI/ML research", "personal health", "startup intelligence"]

## Topics
[List research topics. Each gets a subdirectory under `topics/`.]
- topic-slug: [description]
- topic-slug: [description]

## Topic Grouping Rules
- Each research topic = `topics/{topic-slug}/` with its own `concepts/`, `entities/`, `comparisons/`, `index.md`
- Only appears in one topic → place in that topic's subdirectory
- Appears in 2+ topics → place in global `concepts/` or `entities/`
- When unsure, check existing pages across topics before placing
- Cross-topic pages should link back to all relevant topics via `[[wikilink]]`

## Conventions
- **Markdown formatting:** Wiki pages use Obsidian Flavored Markdown. For wikilinks, embeds, callouts, frontmatter properties, and other Obsidian-specific syntax, refer to the `obsidian-markdown` skill.
- **Academic sources:** For arXiv paper search, Semantic Scholar citations, and BibTeX generation, refer to the `arxiv` skill.
- **Note discipline:** Every page write runs the full seven-item Zettelkasten change checklist (atomicity, split, duplicate, links, structure/index, self-contained, own words) — refer to the `zettelkasten` skill for split criteria and linking rules.
- File names: title as filename — English content: lowercase, hyphens, no spaces (e.g., `transformer-architecture.md`); Chinese content: use the Chinese title directly (e.g., `間隔重複比集中複習有效.md`), per the `zettelkasten` skill's naming convention
- Every wiki page starts with YAML frontmatter (see below)
- Use `[[wikilinks]]` to link between pages (at least 1-2 real outbound links per page — link where the text mentions them, don't pad to hit a quota)
- When updating a page, always bump the `updated` date
- Every new page must be added to `index.md` under the correct section
- Every action must be appended to `log.md`
- **Provenance markers:** On pages that synthesize 3+ sources, append `^[raw/articles/source-file.md]`
  at the end of paragraphs whose claims come from a specific source. This lets a reader trace each
  claim back without re-reading the whole raw file. Optional on single-source pages where the
  `sources:` frontmatter is enough.

## Frontmatter
  ```yaml
  ---
  title: Page Title
  created: YYYY-MM-DD
  updated: YYYY-MM-DD
  type: entity | concept | comparison | query
  tags: [from taxonomy below]
  sources: [raw/articles/source-name.md]
  # Optional quality signals:
  confidence: high | medium | low        # how well-supported the claims are
  contested: true                        # set when the page has unresolved contradictions
  contradictions: [other-page-slug]      # pages this one conflicts with
  ---

confidence and contested are optional but recommended for opinion-heavy or fast-moving topics. Lint surfaces contested: true and confidence: low pages for review so weak claims don't silently harden into accepted wiki fact.

Type governance: adding a new type value requires three things at once — a one-line definition, a title convention, and a lint mapping. Never add a bare enum value.

raw/ Frontmatter

Raw sources ALSO get a small frontmatter block so re-ingests can detect drift:

---
source_url: https://example.com/article   # original URL, if applicable
ingested: YYYY-MM-DD
sha256: <hex digest of the raw content below the frontmatter>
---

The sha256: lets a future re-ingest of the same URL skip processing when content is unchanged, and flag drift when it has changed. Compute over the body only (everything after the closing ---), not the frontmatter itself.

Tag Taxonomy

[Define 10-20 top-level tags for the domain. Add new tags here BEFORE using them.]

Example for AI/ML:

• Models: model, architecture, benchmark, training
• People/Orgs: person, company, lab, open-source
• Techniques: optimization, fine-tuning, inference, alignment, data
• Meta: comparison, timeline, controversy, prediction

Rule: every tag on a page must appear in this taxonomy. If a new tag is needed, add it here first, then use it. This prevents tag sprawl.

Page Thresholds

• Create a page when an entity/concept appears in 2+ sources OR is central to one source
• Add to existing page when a source mentions something already covered
• DON'T create a page for passing mentions, minor details, or things outside the domain
• Promote synthesis-born ideas — when a comparison/query page produces a claim that stands on its own (a mechanism, a generalizable conclusion), create a concept page for it and link it in place; wiki-internal origin counts toward page creation, not just source appearances
• Split a page when it covers 2+ distinct ideas (Zettelkasten atomicity — see the zettelkasten skill's split criteria) — break into atomic pages with cross-links. Length is a warning, not a criterion: run the split criteria on pages over ~200 lines, but a long single-idea page stays whole. Comparison, query, and index pages are exempt from atomicity
• Archive a page when its content is fully superseded — move to _archive/, remove from index

Entity Pages

One page per notable entity. Include:

• Overview / what it is
• Key facts and dates
• Relationships to other entities ([[wikilinks]])
• Source references

Concept Pages

One page per concept or topic. Include:

• Definition / explanation
• Current state of knowledge
• Open questions or debates
• Related concepts ([[wikilinks]])

Comparison Pages

Side-by-side analyses. Include:

• What is being compared and why
• Dimensions of comparison (table format preferred)
• Verdict or synthesis
• Sources

Update Policy

When new information conflicts with existing content:

1. Check the dates — newer sources generally supersede older ones
2. If genuinely contradictory, note both positions with dates and sources
3. Mark the contradiction in frontmatter: contradictions: [page-name]
4. Flag for user review in the lint report
5. When a contradiction is resolved (user ruling or a decisive source): update both pages, clear contested, remove the pair from contradictions:, archive or mark the superseded claim, and log the resolution

### index.md Template

Two levels: global index (wiki root) and topic index (each topic subdirectory).

**Global index.md:**

```markdown
# Wiki Index

> Global catalog. Research topics and cross-topic pages.
> Last updated: YYYY-MM-DD | Topics: N | Total pages: N

## Topics

| Topic | Pages | Index |
|-------|-------|-------|
| [topic name] | N | [[topics/{slug}/index|→ index]] |

## Cross-Topic Concepts

| Page | Summary | Topics |
|------|---------|--------|
| [[page-name]] | One-line summary | topic-a, topic-b |

## Cross-Topic Entities

| Page | Summary | Topics |
|------|---------|--------|

## Comparisons

| Page | Summary |
|------|---------|

## Queries

| Page | Summary |
|------|---------|

Topic index.md (topics/{slug}/index.md):

# {Topic Name} — Index

> Pages: N | Last updated: YYYY-MM-DD

## Concepts

| Page | Summary |
|------|---------|

## Entities

| Page | Summary |
|------|---------|

## Comparisons

| Page | Summary |
|------|---------|

Scaling rule: When any section exceeds 50 entries, split into sub-sections.

log.md Template

# Wiki Log

> Chronological record of all wiki actions. Append-only.
> Format: `## [YYYY-MM-DD] action | subject`
> Actions: ingest, update, query, lint, create, archive, delete
> When this file exceeds 500 entries, rotate: rename to log-YYYY-MM.md (year-month of rotation), start fresh.

## [YYYY-MM-DD] create | Wiki initialized
- Domain: [domain]
- Structure created with SCHEMA.md, index.md, log.md

Core Operations

1. Ingest

When the user provides a source (URL, file, paste), integrate it into the wiki:

① Capture the raw source:

• URL → use defuddle (run npx defuddle <url>) to extract clean markdown; fallback to web_extract if defuddle fails or is unavailable
• PDF → use defuddle (handles PDFs natively); fallback to web_extract
• Pasted text → save to appropriate raw/ subdirectory
• Name the file descriptively: raw/articles/karpathy-llm-wiki-2026.md
• Add raw frontmatter (source_url, ingested, sha256 of the body). On re-ingest of the same URL: recompute the sha256, compare to the stored value — skip if identical, flag drift and update if different. This is cheap enough to do on every re-ingest and catches silent source changes.

② Discuss takeaways with the user — what's interesting, what matters for the domain. (Skip this in automated/cron contexts — proceed directly.)

③ Check what already exists — search index.md and use search_files to find existing pages for mentioned entities/concepts. This is the difference between a growing wiki and a pile of duplicates.

④ Write or update wiki pages:

• Zettelkasten check (every page write): Run the full seven-item change checklist from the zettelkasten skill (atomicity → split → duplicate → links → structure → self-contained → own words). Does the page now cover 2+ distinct ideas → split into linked atomic pages, fix inbound links, update the index.
• New entities/concepts: Create pages only if they meet the Page Thresholds in SCHEMA.md (2+ source mentions, or central to one source)
• Existing pages: Add new information, update facts, bump updated date. When new info contradicts existing content, follow the Update Policy.
• Cross-reference: Every new or updated page must link to at least 1-2 other pages via [[wikilinks]], placed where the text mentions them — real relationships only. Check that existing pages link back.
• Tags: Only use tags from the taxonomy in SCHEMA.md
• Provenance: On pages synthesizing 3+ sources, append ^[raw/articles/source.md] markers to paragraphs whose claims trace to a specific source.
• Confidence: For opinion-heavy, fast-moving, or single-source claims, set confidence: medium or low in frontmatter. Don't mark high unless the claim is well-supported across multiple sources.

⑤ Update navigation:

• Add new pages to index.md under the correct section, alphabetically
• When updating an existing page, refresh its one-line summary in the index if the content meaningfully shifted
• Update the "Total pages" count and "Last updated" date in index header
• Append to log.md: ## [YYYY-MM-DD] ingest | Source Title
• List every file created or updated in the log entry
• Record split_decisions in the log entry: pages split (which criterion hit) and borderline pages judged not-split (why) — this persists the zettelkasten Output Contract into the wiki so the next session can audit what was checked
• Record links_updated: backlinks added, and backlink decisions declined with reasons (lint's symmetry check skips recorded declined pairs)

⑥ Report what changed — list every file created or updated, plus the split_decisions and links_updated records. If the ingest touched 10+ pages, suggest running a light lint subset (broken links, index completeness, log reconciliation) before ending the session.

A single source can trigger updates across 5-15 wiki pages. This is normal and desired — it's the compounding effect.

Academic Sources (arXiv + Semantic Scholar)

For academic papers, use the arxiv skill to search and retrieve:

1. Search: python arxiv/scripts/search_arxiv.py "topic" --sort date --max 10
2. Impact check: Semantic Scholar API for citation counts

   curl -s "https://api.semanticscholar.org/graph/v1/paper/arXiv:ID?fields=citationCount,influentialCitationCount"
   

3. Read: npx defuddle https://arxiv.org/pdf/ID (fallback: web_extract)
4. Ingest as normal — raw source goes to raw/papers/, wiki pages follow standard workflow

On wiki pages for papers, add optional frontmatter fields:

arxiv_id: "2402.03300"
citation_count: 150
influential_citations: 12

2. Query

When the user asks a question about the wiki's domain:

① Read index.md to identify relevant pages. ② For wikis with 100+ pages, also search_files across all .md files for key terms — the index alone may miss relevant content. ③ Read the relevant pages using read_file. ④ Synthesize an answer from the compiled knowledge. Cite the wiki pages you drew from: "Based on [[page-a]] and [[page-b]]..." ⑤ File valuable answers back — if the answer is a substantial comparison, deep dive, or novel synthesis, create a page in queries/ or comparisons/. Don't file trivial lookups — only answers that would be painful to re-derive. ⑥ Update log.md with the query and whether it was filed.

3. Lint

When the user asks to lint, health-check, or audit the wiki (also suggest it proactively after any ingest touching 10+ pages — see ingest ⑥). Items marked (semantic, agent-run) are judgment-based candidate reports, not mechanical checks — expect false positives, report only:

① Orphan pages: Find pages with no inbound [[wikilinks]] from other pages.

# Use execute_code for this — programmatic scan across all wiki pages
import os, re
from collections import defaultdict
wiki = "<wiki>"  # replace with resolved wiki path
# Scan all .md files in entities/, concepts/, comparisons/, queries/
# Extract all [[wikilinks]] — build inbound link map
# Pages with zero inbound links are orphans

② Broken wikilinks: Find [[links]] that point to pages that don't exist.

③ Index completeness: Every wiki page should appear in index.md. Compare the filesystem against index entries.

④ Frontmatter validation: Every wiki page must have all required fields (title, created, updated, type, tags, sources). Tags must be in the taxonomy.

⑤ Stale content: Pages whose updated date is >90 days older than the most recent source that mentions the same entities.

⑥ Contradictions: Pages on the same topic with conflicting claims. Look for pages that share tags/entities but state different facts. Surface all pages with contested: true or contradictions: frontmatter for user review.

⑦ Quality signals: List pages with confidence: low and any page that cites only a single source but has no confidence field set — these are candidates for either finding corroboration or demoting to confidence: medium.

⑧ Source drift: For each file in raw/ with a sha256: frontmatter, recompute the hash and flag mismatches. Mismatches indicate the raw file was edited (shouldn't happen — raw/ is immutable) or ingested from a URL that has since changed. Not a hard error, but worth reporting.

⑨ Page size & atomicity: Flag pages covering 2+ distinct ideas — candidates for splitting per the zettelkasten skill's split criteria. Pages over ~200 lines are a warning to run those criteria, not grounds to split by length alone. Skip index pages. For comparison/query pages, replace the atomicity check with a promotion spot-check (semantic, agent-run): standalone ideas living only inside a synthesis page that were never promoted to concept pages (see Page Thresholds).

⑩ Tag audit: List all tags in use, flag any not in the SCHEMA.md taxonomy.

⑪ Log rotation: If log.md exceeds 500 entries, rotate it.

⑫ Log reconciliation: compare recent page updated dates against log entries — flag pages changed with no log record (writes that bypassed the discipline).

⑬ Link symmetry: build the bidirectional link map; report A→B edges where B doesn't link back, excluding pairs whose declined-backlink decision is recorded in the log (ingest ⑤). Backlinks are curation judgments — report for review, don't auto-fix. Exception: contradictions frontmatter must be mutual (A lists B ⇒ B lists A) — asymmetry there is a hard error.

⑭ Frontmatter link fields: verify every slug in contradictions: (and any other frontmatter field that references pages) resolves to an existing page — renames and archiving leave these dangling silently, since they are not [[wikilinks]] and no other check covers them.

⑮ Misplacement & promote: find topics/{t}/ pages referenced from 2+ topics — candidates for promotion to global. Promote = move the file keeping the filename (filename-based [[wikilinks]] survive), fix any path-style links, update both indexes, log the move.

⑯ Duplicate page detection (semantic, agent-run): near-duplicate titles or high content overlap between pages sharing tags/entities.

⑰ Verbatim-copy check (semantic, agent-run): wiki pages whose paragraphs are near-verbatim from raw/ — entity facts, figures, and provenance markers legitimately repeat, so candidates only.

⑱ Title-form ↔ type spot check (semantic, agent-run): sample pages — concept pages should carry statement titles, entity pages name titles.

⑲ Index summary drift: sample index rows against current page content — flag stale one-line summaries.

⑳ Report findings with specific file paths and suggested actions, grouped by severity (broken links > orphans > dangling frontmatter slugs > source drift > contested pages > stale content > style issues).

㉑ Append to log.md: ## [YYYY-MM-DD] lint | N issues found

Working with the Wiki

Searching

# Find pages by content (replace <wiki> with resolved path)
search_files "transformer" path="<wiki>" file_glob="*.md"

# Find pages by filename
search_files "*.md" target="files" path="<wiki>"

# Find pages by tag
search_files "tags:.*alignment" path="<wiki>" file_glob="*.md"

# Recent activity
read_file "<wiki>/log.md" offset=<last 20 lines>

Bulk Ingest

When ingesting multiple sources at once, batch the updates:

1. Read all sources first
2. Identify all entities and concepts across all sources
3. Check existing pages for all of them (one search pass, not N)
4. Create/update pages in one pass (avoids redundant updates)
5. Update index.md once at the end
6. Append log entries per batch as you go — don't hold them all for the end (removes the interruption window); include per-page split_decisions and links_updated
7. Defer backlink decisions to a single pass at the end of the batch (one pass, not N — same principle as step 3)

Archiving

When content is fully superseded or the domain scope changes:

1. Create _archive/ directory if it doesn't exist
2. Move the page to _archive/ with its original path (e.g., _archive/entities/old-page.md)
3. Remove from index.md
4. Update any pages that linked to it — replace wikilink with plain text + "(archived)"; also remove or repoint its slug in any contradictions: or other frontmatter fields that reference it
5. Log the archive action

Obsidian Integration

The wiki directory works as an Obsidian vault out of the box:

• [[wikilinks]] render as clickable links
• Graph View visualizes the knowledge network
• YAML frontmatter powers Dataview queries
• The raw/assets/ folder holds images referenced via ![[image.png]]

For best results:

• Set Obsidian's attachment folder to raw/assets/
• Enable "Wikilinks" in Obsidian settings (usually on by default)
• Install Dataview plugin for queries like TABLE tags WHERE type = "entity" AND contains(tags, "company") — query by frontmatter type, not by folder: FROM "entities" would miss topics/*/entities/

If using the Obsidian skill alongside this one, set OBSIDIAN_VAULT_PATH to the same directory as the wiki path.

Obsidian Headless (servers and headless machines)

On machines without a display, use obsidian-headless instead of the desktop app. It syncs vaults via Obsidian Sync without a GUI — perfect for agents running on servers that write to the wiki while Obsidian desktop reads it on another device.

Setup:

# Requires Node.js 22+
npm install -g obsidian-headless

# Login (requires Obsidian account with Sync subscription)
ob login --email <email> --password '<password>'

# Create a remote vault for the wiki
ob sync-create-remote --name "LLM Wiki"

# Connect the wiki directory to the vault
cd <wiki>
ob sync-setup --vault "<vault-id>"

# Initial sync
ob sync

# Continuous sync (foreground — use systemd for background)
ob sync --continuous

Continuous background sync via systemd:

# ~/.config/systemd/user/obsidian-wiki-sync.service
[Unit]
Description=Obsidian LLM Wiki Sync
After=network-online.target
Wants=network-online.target

[Service]
ExecStart=/path/to/ob sync --continuous
WorkingDirectory=<wiki>
Restart=on-failure
RestartSec=10

[Install]
WantedBy=default.target

systemctl --user daemon-reload
systemctl --user enable --now obsidian-wiki-sync
# Enable linger so sync survives logout:
sudo loginctl enable-linger $USER

This lets the agent write to the wiki on a server while you browse the same vault in Obsidian on your laptop/phone — changes appear within seconds.

Pitfalls

• Never modify files in raw/ — sources are immutable. Corrections go in wiki pages.
• Always orient first — read SCHEMA + index + recent log before any operation in a new session. Skipping this causes duplicates and missed cross-references.
• Always update index.md and log.md — skipping this makes the wiki degrade. These are the navigational backbone.
• Don't create pages for passing mentions — follow the Page Thresholds in SCHEMA.md. A name appearing once in a footnote doesn't warrant an entity page.
• Don't create pages without cross-references — isolated pages are invisible. Every page must link to at least 1-2 other pages, via real relationships rather than quota-padding.
• Frontmatter is required — it enables search, filtering, and staleness detection.
• Tags must come from the taxonomy — freeform tags decay into noise. Add new tags to SCHEMA.md first, then use them.
• Keep pages scannable — a wiki page should be readable in 30 seconds. Split pages that cover 2+ distinct ideas; length (~200 lines) is a warning to run the split criteria, not a reason to split by itself. Move detailed analysis to dedicated deep-dive pages.
• Run the Zettelkasten checklist on every page write — atomicity and split decisions happen at write time, not only during lint. See the zettelkasten skill.
• Ask before mass-updating — if an ingest would touch 10+ existing pages, confirm the scope with the user first.
• Rotate the log — when log.md exceeds 500 entries, rename it log-YYYY-MM.md and start fresh. The agent should check log size during lint.
• Handle contradictions explicitly — don't silently overwrite. Note both claims with dates, mark in frontmatter, flag for user review.

Related Tools

llm-wiki-compiler is a Node.js CLI that compiles sources into a concept wiki with the same Karpathy inspiration. It's Obsidian-compatible, so users who want a scheduled/CLI-driven compile pipeline can point it at the same vault this skill maintains. Trade-offs: it owns page generation (replaces the agent's judgment on page creation) and is tuned for small corpora. Use this skill when you want agent-in-the-loop curation; use llmwiki when you want batch compile of a source directory.