LLM Wiki Manager

A skill for running a personal LLM-managed wiki: a persistent, compounding markdown knowledge base where the human curates sources and asks questions, and the LLM does all the reading, writing, cross-referencing, and bookkeeping.

The pattern is intentionally tool-agnostic — it's just markdown files in a git repo. It works in any editor (Obsidian, VS Code, plain vim), and the skill makes no assumptions beyond "you have a filesystem and git". The user is on Claude Code.

The three layers (memorize this)

Every wiki this skill manages has exactly three layers. Confusing them is the most common failure mode.

1. raw/ — immutable sources the user curates. Articles, PDFs, transcripts, screenshots, data files. The LLM reads from raw/ but never modifies it. This is the source of truth; if it gets edited the audit trail breaks. 2. wiki/ — markdown pages the LLM owns entirely. Source summaries, entity pages, concept pages, comparisons. Three structural files always live here: wiki/index.md (content catalog), wiki/log.md (operation log), wiki/hot.md (~500-word hot cache, rewritten after every ingest). The user reads it; the LLM writes it. Pages link to each other and back to specific files in raw/. 3. CLAUDE.md (at the wiki root) — the schema. Tells future LLM sessions how this particular wiki is structured, what the page conventions are, how categories are named, how aggressive ingest should be. Co-evolved with the user over time — when something works, write it down here so the next session inherits it.

The whole point of the pattern is that wiki/ is a persistent, compounding artifact. It is not re-derived on every query. New sources update existing pages in place. That is what separates this from RAG.

Mode router

When this skill triggers, identify which mode applies and read the matching reference file. The user rarely names the mode explicitly — infer it from context. If genuinely ambiguous, ask one short clarifying question.

If multiple modes apply (e.g., user asks a question and wants the answer filed back), do them in sequence and log each one.

Step 0 — schema version check (every mode). The wiki's CLAUDE.md frontmatter carries a schema_version stamp (unstamped = v1). If it is lower than the version this skill expects, offer one sentence before starting the requested work: "This wiki is at schema v1, the skill expects v2 — run scripts/migrate_wiki.py first?" Don't block the requested operation if the user declines; lint also reports the mismatch as a finding.

Update vs. Ingest's Disputes handling — important distinction. Ingest already handles contradictions on a single page by adding a Disputes section. Switch to Update mode only when the new source's claim affects multiple existing pages — that is, when the same idea is paraphrased across the wiki and a single Disputes section won't keep the wiki internally consistent. If only one page is affected, stay in ingest. See references/update-workflow.md for the precise trigger checklist.

Multi-wiki requires CLAUDE.md configuration. Before any multi-wiki operation, read the project CLAUDE.md for an ## External Wiki section that declares the global wiki path and routing rules. If the section is missing and a multi-wiki operation is requested, run Schema-evolve first to add it, then proceed. Never guess or assume the global wiki path. See references/multi-wiki-routing.md for the one-time setup block and the four canonical scenarios.

Core invariants (apply in every mode) — nine rules

These are the disciplines that make a wiki compound. Skipping them is how wikis rot.

1. The user owns raw/. The LLM owns wiki/.

Never write to raw/. Never ask the user to edit wiki/ by hand (they can if they want, but the skill never asks). When summarizing a source, the summary lives in wiki/sources/<source-slug>.md and links back to the file in raw/.

2. Every operation logs to wiki/log.md.

After any ingest, query, update, or lint, append a single entry using scripts/append_log.py --path <wiki-root>:

## [YYYY-MM-DD] ingest | <source title>
## [YYYY-MM-DD] query  | <one-line question>
## [YYYY-MM-DD] update | <claim corrected, source authority>
## [YYYY-MM-DD] lint   | <pages touched or "clean">

Consistent prefix matters — it makes the log greppable: grep "^## \[" wiki/log.md | tail -20 shows recent activity. The script enforces the format. For lint, the script auto-logs by default — you don't need a separate append_log.py call after running lint_wiki.py. For ingest/query/update, the LLM calls the script.

3. Every new or significantly-updated page touches wiki/index.md.

Use scripts/update_index.py --path <wiki-root>. The index is content-oriented: category, title, one-line summary, link. It is the LLM's primary navigation aid in future sessions when the wiki is large. Stale index = wiki that feels lost.

One page = one index entry. A page never appears under multiple headings — grouping is by theme/category, not by tag (tag-grouped indexes duplicate every multi-tag page). Hub pages are marked with ★ after the link. The script warns when a page is about to be indexed in a second category.

4. Cross-reference aggressively.

When ingesting a source about, say, a person who already has an entity page, update that entity page with the new information. Don't leave the connection implicit in the source summary. The whole value proposition is that connections are made eagerly, while context is fresh, not lazily at query time.

Use markdown wiki-style links: [Page Title](../entities/page-title.md) or, if the user prefers Obsidian-style, [[page-title]]. Pick one convention and record it in CLAUDE.md.

5. Cite back to raw/.

Every claim in a wiki page should be traceable to a specific source. The standard pattern in source-summary pages is:

> Quote or paraphrase from the source.
> — `raw/2026-01-pollan-eat-food.pdf`, p. 14

In synthesis pages (entity, concept, comparison), use inline references like (see [Pollan 2026](../sources/pollan-eat-food.md)). The reader should always be able to trace a claim back to a real source.

6. Flag contradictions, don't silently overwrite.

If a new source contradicts an existing claim on a wiki page, do not just replace the old claim. Add the new claim alongside, mark both with their source, and note the conflict — usually as a > [!warning] Sources disagree callout or a "Disputes" subsection. The user decides which to believe; the wiki records the disagreement.

7. Rewrite wiki/hot.md after every ingest.

wiki/hot.md is a ~500-word hot cache with exactly five sections: State, Last ingest, Active themes, Open items, Conventions (see assets/templates/hot.md.tmpl). It is the first file a cross-project reader sees. Rewrite entirely — do not append. Dated changelog blocks belong in log.md, never in hot. The tag inventory does NOT live here — the canonical tag list lives in the wiki's CLAUDE.md. See references/ingest-workflow.md Step 10 for the rewrite checklist; lint_wiki.py flags hot.md bloat.

8. The schema lives in CLAUDE.md and is updated when conventions change.

If a working pattern emerges in conversation ("let's always tag book chapters with chapter:N"), write it into CLAUDE.md immediately so the next session inherits it. The user shouldn't have to re-explain conventions every session.

9. If the project's CLAUDE.md declares an External Wiki, honor it.

Some users run two wikis: a per-project wiki at the working directory and a long-lived global wiki (often an existing Obsidian vault) at a fixed path. The project's CLAUDE.md declares the global wiki's location with a section like:

## External Wiki

Global knowledge base: ~/Documents/obsidian/

When this section is present, read both CLAUDE.md files at session start (project + global) so you know each wiki's schema. Route writes per the user's rules: project-specific knowledge to the project wiki, portable concepts to the global wiki. Always pass --path to scripts targeting the chosen wiki — never assume the current working directory is the right target. Cross-wiki links must be absolute (~/...), never relative across wiki boundaries. See references/multi-wiki-routing.md for the four canonical scenarios (write-to-global, pull-from-global, promote, dual-lint).

Standard wiki layout

init_wiki.py produces:

<wiki-root>/
├── CLAUDE.md          # Schema for the wiki itself (instructions to future Claude)
├── README.md          # Human-facing overview (optional but useful)
├── raw/               # User-curated sources — LLM reads, never writes
│   └── .gitkeep
└── wiki/              # LLM-managed pages
    ├── index.md       # Content catalog by category
    ├── log.md         # Chronological log of operations
    ├── hot.md         # ~500-word hot cache — rewrite after every ingest
    ├── sources/       # One summary page per ingested source
    ├── entities/      # People, organizations, products, places
    ├── concepts/      # Ideas, theories, frameworks, terms
    ├── notes/         # Loose pages that don't fit elsewhere yet
    └── reports/       # Auto-generated dated lint/audit reports (scripts write here, not the LLM)

The entities/, concepts/, sources/, notes/ split is the default suggestion. Many wikis converge on a different taxonomy as they grow (e.g. a research wiki might use papers/, methods/, findings/, open-questions/). When the natural taxonomy diverges, update CLAUDE.md to reflect the new convention and reorganize incrementally.

reports/ is special: it's automated output, not user content. lint_wiki.py (and future audit tooling) writes dated files here — lint-2026-05-07.md, audit-saturated-fat-2026-05-07.md — and same-day re-runs overwrite (idempotent daily). Lint excludes this folder from orphan/stub/index-drift checks.

Bundled scripts

All five are in scripts/. They are deliberately small, idempotent, and dependency-free (Python stdlib only).

• init_wiki.py — scaffold a new wiki. Creates wiki/index.md, wiki/log.md, wiki/hot.md, categorical subdirs, CLAUDE.md, README.md. On an existing wiki (already has .md files in wiki/) only adds wiki/reports/ if missing — never imposes categorical dirs or overwrites files. Idempotent. See references/bootstrap-workflow.md.
• append_log.py — append a log entry with the consistent ## [YYYY-MM-DD] action | title prefix. Use after every ingest, query, update, and lint. Auto-detects wiki/log.md (standard) or root/log.md (flat layout). Always pass --path <wiki-root>. Valid actions: ingest, query, update, lint, audit, bootstrap, schema-evolve, note, setup, restructure.
• update_index.py — add or update an entry under a category in wiki/index.md. Auto-detects wiki/index.md (standard) or root/index.md (flat layout). Preserves #tag-style category names. Always pass --path <wiki-root> and --page-path <path-to-page>. Idempotent on (category, title).
• lint_wiki.py — scan for orphan pages, broken links (markdown and [[wiki-links]]), index/filesystem drift, duplicate index entries, hot.md bloat, tag hygiene (single-use tags, over-tagged pages), ambiguous wiki-link slugs, log gaps, and schema-version drift. wiki/hot.md excluded from orphan/stub/index checks but gets its own health check. Auto-detects log and index in both wiki/ and root layouts. Default behavior: writes report to wiki/reports/lint-<today>.md, then auto-tracks (adds Reports index entry, appends lint | Health check log entry). Same-day re-runs overwrite the report file. Override with --stdout, --report PATH, or --no-track; thresholds via --stub-words, --log-gap-days, --hot-max-words, --max-tags. See references/lint-workflow.md.
• migrate_wiki.py — upgrade an existing wiki to the current structural conventions when a skill update changes them. Reads the schema_version stamp in the wiki's CLAUDE.md (unstamped = v1), shows a dry-run of the mechanical steps plus the manual (LLM) steps, applies with --apply only. Never touches raw/; logs every step as restructure; idempotent (re-running at the current version is a no-op). See references/migrate-workflow.md.

Always prefer running these scripts to hand-editing wiki/index.md or wiki/log.md. They keep the format consistent across sessions and make the wiki greppable.

For update mode, no new script is needed — the workflow uses append_log.py (with --action update) and update_index.py orchestrated by the LLM following references/update-workflow.md. The diffs and multi-page sweep are LLM work, not script work, because they require semantic judgment.

For multi-wiki mode, scripts are called with --path <wiki-root> to operate on the chosen wiki. The routing decision (which wiki receives the write) comes from the project CLAUDE.md's ## External Wiki section. Never rely on the current working directory — always pass --path explicitly. See references/multi-wiki-routing.md.

Bundled templates

In assets/templates/. Use them as starting points; adapt freely:

• wiki-CLAUDE.md.tmpl — the schema file dropped into a fresh wiki by init_wiki.py
• index.md.tmpl, log.md.tmpl — initial scaffolding for wiki/index.md and wiki/log.md
• hot.md.tmpl — initial scaffolding for wiki/hot.md (five fixed sections: State, Last ingest, Active themes, Open items, Conventions)
• source-summary.md.tmpl — one ingested source, full structure
• entity-page.md.tmpl — a person, org, place, or product
• concept-page.md.tmpl — an idea, framework, or term
• comparison-page.md.tmpl — for "X vs Y" pages, often produced as query outputs

assets/examples/healthy-wiki-tree.txt shows what a mature wiki of ~50 sources looks like — useful when teaching the pattern or sanity-checking organization.

When in doubt

Read references/philosophy.md for the why and references/architecture.md for the what. Both are short. The workflow files are the how: bootstrap-workflow.md, ingest-workflow.md, query-workflow.md, update-workflow.md, multi-wiki-routing.md, lint-workflow.md, migrate-workflow.md, schema-design-guide.md, teaching-mode.md.

The single most important principle: the LLM does the bookkeeping. Cross-references, index updates, log entries, contradiction-flagging, stale-claim review, multi-wiki routing. That work is what makes the wiki compound. Skipping it because "the user didn't ask" is the failure mode this skill exists to prevent.