Internalized Wiki

Build and maintain a personal LLM wiki for Obsidian. The agent owns source ingestion and compiled knowledge articles; the user owns internalization notes and review readiness.

Core ideas:

• Sources are preserved under raw/.
• Durable synthesized knowledge compounds under wiki/.
• Obsidian YAML tracks internalization and review state.
• The ## 内化 section is user-authored and must be preserved.

Architecture

All paths are under the user's project root:

raw/ - Immutable source material. Read existing raw files but do not rewrite them. Organize by topic subdirectories such as raw/machine-learning/.

wiki/ - Compiled knowledge articles. The agent may create and update these files, while preserving user-owned internalization content. Organize one topic level deep: wiki/<topic>/<article>.md.

wiki/index.md - Global index. One row per article, grouped by topic, with link, summary, and Updated date.

wiki/log.md - Append-only operation log.

Templates live in references/ relative to this file. Read them when exact formatting is needed:

• references/raw-template.md
• references/article-template.md
• references/archive-template.md
• references/index-template.md

Initialization

Initialize only on the first ingest. Check whether raw/ and wiki/ exist. Create only what is missing; never overwrite existing files:

• raw/ directory with .gitkeep
• wiki/ directory with .gitkeep
• wiki/index.md with heading # Knowledge Base Index
• wiki/log.md with heading # Wiki Log

If query or lint cannot find the wiki structure, tell the user: "Run an ingest first to initialize the wiki." Do not auto-create it.

Metadata Model

Normal wiki articles use YAML frontmatter for Obsidian and review automation:

---
内化: false
review:
  state: new
  due: null
  last_reviewed: null
  difficulty: null
  stability: null
  retrievability: null
---

Keep human-readable provenance immediately after the title:

> Sources: ...
> Raw: ...

Use YAML for machine-readable internalization and review state. Use blockquotes for source traceability. Do not replace one with the other.

Ingest

Fetch source material into raw/, then compile it into wiki/. Always do both steps.

Fetch Raw Source

1. Get the source content with available web or file tools. If no tool can reach it, ask the user to paste it.
2. Pick a topic directory. Check existing raw/ subdirectories first; reuse a close topic. Create a new topic only for genuinely distinct material.
3. Save as raw/<topic>/YYYY-MM-DD-descriptive-slug.md.

- Build the slug from the source title, kebab-case, max 60 characters. - If published date is unknown, omit the date prefix from the filename and set metadata Published to Unknown. - If the filename exists, append a numeric suffix such as -2. - Include source URL or origin, collected date, and published date. - Preserve source text faithfully. Clean formatting noise but do not rewrite opinions or meaning.

Use references/raw-template.md for exact format.

Compile Wiki Knowledge

Determine where the source belongs:

• Same core thesis as an existing article: merge into that article.
• New concept: create a new article in the most relevant topic directory.
• Spans multiple topics: place it in the most relevant topic and add See Also links.

A single source may both update an existing article and create another article if it introduces a distinct concept.

When creating a new article, use references/article-template.md.

When updating an existing article:

• Preserve any user-written ## 内化 section exactly as-is.
• If YAML frontmatter is missing, add the default metadata block.
• If YAML frontmatter exists, merge only required article metadata changes.
• Never overwrite non-null review fields. Those belong to the review system.
• Refresh the Updated date when article knowledge changes.

Check for factual conflicts. If a new source contradicts existing content, annotate the disagreement with source attribution. If conflicting content lives in separate articles, note it in both and cross-link them.

Cascade Updates

After the primary article:

1. Scan articles in the same topic directory for materially affected content.
2. Scan wiki/index.md entries in other topics for related concepts.
3. Update every article whose knowledge is materially affected.

Archive pages are point-in-time snapshots and are never cascade-updated.

Post-Ingest

Update wiki/index.md for every touched article. When adding a topic section, include a one-line description. Updated dates reflect when article knowledge changed, not filesystem timestamps.

Append to wiki/log.md:

## [YYYY-MM-DD] ingest | <primary article title>
- Updated: <cascade-updated article title>
- Updated: <another cascade-updated article title>

Omit - Updated: lines when no cascade updates occurred.

Internalization Workflow

The agent writes or updates compiled knowledge. The user reads the article in Obsidian, writes personal understanding under ## 内化, then changes YAML 内化 from false to true when the note is ready for review.

Treat ## 内化 as user-authored content:

• The agent may read it during ingest, query, or lint.
• The agent must not overwrite, rewrite, summarize, relocate, or delete it.
• Template comments or placeholders under ## 内化 do not count as user-written internalization.
• If 内化: true, ## 内化 must exist and contain non-placeholder content to be review-ready.

Query

Search the wiki and answer questions. Triggers include:

• "What do I know about X?"
• "Summarize everything related to Y"
• "Compare A and B based on my wiki"

Steps:

1. Read wiki/index.md to locate relevant articles.
2. Read those articles and synthesize an answer.
3. Prefer wiki content over training knowledge.
4. Cite sources with project-root-relative markdown links, such as [Article Title](wiki/topic/article.md).
5. Output the answer in conversation. Do not write files unless asked.

Archiving

When the user explicitly asks to archive or save a query answer to the wiki:

1. Write the answer as a new wiki page using references/archive-template.md.

- Convert conversation citations to paths relative to the archive file. - Use Sources links to cited wiki articles. - Do not include a Raw field because archive content comes from synthesized wiki answers. - Name the file after the query topic. - Place it in the most relevant topic directory.

1. Always create a new page. Never merge archive content into normal articles.
2. Update wiki/index.md. Prefix the Summary with [Archived].
3. Append to wiki/log.md:

## [YYYY-MM-DD] query | Archived: <page title>

Archive pages may keep YAML frontmatter if present and must preserve any existing user-written ## 内化 content if they are later touched.

Lint

Run quality checks on the wiki. Deterministic checks may be auto-fixed. Heuristic checks are report-only.

Deterministic Checks

Auto-fix these when safe:

Index consistency

• File exists but is missing from wiki/index.md: add an entry with (no summary) placeholder. For Updated, use article metadata Updated date if present; otherwise use file last modified date.
• Index entry points to a missing file: mark it [MISSING] in the index. Do not delete it.

Internal links

• For markdown links in wiki article bodies and Sources metadata, excluding Raw field links and excluding index/log files: if the target is missing, search wiki/ for a same-named file.
• Exactly one match: fix the path.
• Zero or multiple matches: report to the user.

Raw references

• Every Raw field link must point to an existing raw/ file.
• Missing target with exactly one same-named raw match: fix the path.
• Zero or multiple matches: report to the user.

See Also

• Within each topic directory, add obvious missing cross-references between related articles.
• Remove links to deleted files.

Internalization metadata

• If YAML frontmatter is missing on a normal article, add the default internalization and review block.
• Do not overwrite non-null review fields.
• Preserve ## 内化 content exactly.

Heuristic Checks

Report these without auto-fixing:

• Factual contradictions across articles.
• Outdated claims superseded by newer sources.
• Missing conflict annotations where sources disagree.
• Orphan pages with no inbound links from other wiki articles.
• Missing cross-topic references.
• Concepts frequently mentioned but lacking a dedicated page.
• Archive pages whose cited source articles have been substantially updated since archival.
• 内化: true articles with missing or empty ## 内化.
• ## 内化 sections that contain only comments, placeholders, or template text.

Post-Lint

Append to wiki/log.md:

## [YYYY-MM-DD] lint | <N> issues found, <M> auto-fixed

Conventions

• Use standard markdown with relative links in wiki files.
• In conversation, cite project-root-relative paths.
• Keep wiki/ one topic level deep. No deeper nesting.
• Keep raw/ immutable after source capture.
• Use today's date for log entries, collected dates, and archived dates.
• Published dates come from the source; use Unknown when unavailable.
• Updated dates reflect when article knowledge content changed.
• Ingest, archive, and lint update wiki/log.md.
• Ingest and archive update wiki/index.md; lint updates the index only for deterministic index fixes.
• Plain queries do not write files.

Common Mistakes