Cross-Linker — Automated Wiki Cross-Referencing

You are weaving the wiki's knowledge graph tighter by finding and inserting missing [[wikilinks]] between pages that should reference each other but currently don't.

Follow the Retrieval Primitives table in llm-wiki/SKILL.md. Build the registry in Step 1 by grepping frontmatter only (not full pages). Reserve full Read for the unlinked-mention detection pass, and even there, only read pages whose summaries/titles make them plausible link targets. Blind full-vault reads are what this framework exists to avoid.

Before You Start

Writing profile: Before drafting or rewriting natural-language Markdown, read and apply the Writing Profile Resolution section in llm-wiki/SKILL.md. Framework schema, provenance, safety, and operation-specific requirements take precedence. WRITING.md preferences apply only to newly drafted or rewritten natural-language Markdown; preserve source content and structured records.

1. Resolve config — follow the Config Resolution Protocol in llm-wiki/SKILL.md (inline @name override → walk up CWD for .env → global config → prompt setup). This gives OBSIDIAN_VAULT_PATH and OBSIDIAN_LINK_FORMAT (default: wikilink).
2. Read index.md to get the full inventory of pages and their one-line descriptions
3. Skim log.md to see what was recently ingested (focus linking effort on new pages)

When inserting links in Step 4, apply the link format from llm-wiki/SKILL.md (Link Format section) using the OBSIDIAN_LINK_FORMAT value. When OBSIDIAN_LINK_FORMAT=markdown, compute the relative .md path from the file being edited to the target page.

Step 1: Build the Page Registry

Glob all .md files in the vault (excluding _archives/, _readouts/, .obsidian/). For each page, extract:

• Filename (without .md) — this is the wikilink target
• Title from frontmatter
• Aliases from frontmatter (if any)
• Tags from frontmatter
• Category from frontmatter or directory inference
• One-line summary — first sentence or title field

Build a lookup table:

page_name → { path, title, aliases, tags, summary }

This is your "vocabulary" — every entry in this table is a valid wikilink target.

Step 2: Scan for Missing Links

For each page in the vault:

1. Read the full content
2. Extract existing wikilinks — find all [[...]] references already present
3. Search for unlinked mentions — check if the page's text contains any of these, without being wrapped in [[...]]:
   • Page filenames (e.g., the word "MyProject" appears but [[projects/my-project/my-project]] is missing)
   • Page titles from frontmatter
   • Aliases from frontmatter
   • Entity names, project names, concept names from the registry
4. Check for semantic connections — pages that share multiple tags or are in the same project directory but don't link to each other

Matching Rules

• Case-insensitive matching for names (e.g., "my-project" matches page MyProject)
• Diacritic-insensitive matching — normalize both the page name and the body text with Unicode NFKD (decompose accented characters to base + combining marks, strip combining marks) before comparing. This ensures body text "Muller" matches page [[entities/müller]] and vice versa.
• Skip self-references — a page shouldn't link to itself
• Skip common words — don't link "the", "and", generic terms. Only match on distinctive names
• Prefer the shortest unambiguous wikilink path — use [[page-name]] not [[full/path/to/page-name]] when the name is unique across the vault
• Don't link inside code blocks or frontmatter
• Don't double-link — if [[foo]] already appears on the page, don't add another

Step 3: Score and Rank Suggestions

Not every possible link is worth adding. Score each candidate using a composite signal, then tag it with a confidence label.

Scoring

Confidence labels

Tag each candidate with a confidence label based on its score:

Only act on EXTRACTED and INFERRED candidates. Include the confidence label in the Cross-Link Report so the user can review INFERRED links before trusting them.

Step 4: Apply Links

Pre-write snapshot — before the first file write, check whether the vault itself is the root of a Git repository. Merely being a subdirectory of a larger repository does not qualify: running git add -A there could capture unrelated files. If the vault is not a standalone Git repository, skip this step silently — no nagging, no suggesting git init.

VAULT_REAL_PATH=$(cd "$OBSIDIAN_VAULT_PATH" && pwd -P)
VAULT_GIT_ROOT=$(git -C "$OBSIDIAN_VAULT_PATH" rev-parse --show-toplevel 2>/dev/null || true)
SNAPSHOT_SHA=""

if [ -n "$VAULT_GIT_ROOT" ] && [ "$VAULT_GIT_ROOT" = "$VAULT_REAL_PATH" ]; then
  if git -C "$OBSIDIAN_VAULT_PATH" diff --quiet \
    && git -C "$OBSIDIAN_VAULT_PATH" diff --cached --quiet \
    && [ -z "$(git -C "$OBSIDIAN_VAULT_PATH" ls-files --others --exclude-standard)" ]; then
    SNAPSHOT_SHA=$(git -C "$OBSIDIAN_VAULT_PATH" rev-parse HEAD)
  else
    if ! git -C "$OBSIDIAN_VAULT_PATH" add -A; then
      echo "Pre-write snapshot failed; abort the skill without writing any vault files." >&2
      exit 1
    fi
    if ! git -C "$OBSIDIAN_VAULT_PATH" commit -m "pre-cross-linker snapshot" --quiet; then
      echo "Pre-write snapshot failed; abort the skill without writing any vault files." >&2
      exit 1
    fi
    SNAPSHOT_SHA=$(git -C "$OBSIDIAN_VAULT_PATH" rev-parse HEAD)
  fi
fi

The clean-repository branch deliberately avoids calling git commit, so "nothing to commit" is not treated as an error. If git add or git commit fails, stop before editing the vault; never continue without the promised snapshot.

If SNAPSHOT_SHA is non-empty and the skill writes files, include the SHA in the final report. To discard the entire run, after confirming there are no later changes worth keeping, the user can run:

git -C "$OBSIDIAN_VAULT_PATH" reset --hard "$SNAPSHOT_SHA"
git -C "$OBSIDIAN_VAULT_PATH" clean -fd

For each page with missing links:

4a: Inline linking (preferred)

Find the first natural mention of the term in the body text and wrap it in wikilinks:

Before:

This project uses knowledge graphs to connect entities.

After:

This project uses [[concepts/knowledge-graphs|knowledge graphs]] to connect entities.

Use the [[path|display text]] format when the wikilink path differs from the display text.

4b: Related section (fallback)

If the term isn't mentioned naturally in the body but the pages are semantically related (shared tags, same project), add a ## Related section at the bottom of the page:

## Related

- [[projects/my-project/my-project]] — Also uses AI agents for research automation
- [[concepts/knowledge-graphs]] — Core technique used in this project

If a ## Related section already exists, append to it. Don't duplicate existing entries.

4c: Infer and write relationship type

For every EXTRACTED or INFERRED link added (inline or related section), infer a semantic relationship type from the surrounding sentence context and write it to the page's relationships: frontmatter block. Skip AMBIGUOUS links.

Type inference rules — scan the sentence containing the mention (or, for related-section links, the page title and shared-tag context):

If the surrounding context is ambiguous or the link came from shared-tag matching (no in-body mention), default to related_to.

Writing the block:

Read the page's YAML frontmatter. If a relationships: block already exists, append new entries without duplicating existing targets. If the block is absent, add it after aliases: (or after tags: when aliases: is missing).

relationships:
  - target: "[[concepts/knowledge-graphs]]"
    type: uses

Always use wikilink format ([[path/to/page]]) for target values in the relationships: YAML block — regardless of OBSIDIAN_LINK_FORMAT. The OBSIDIAN_LINK_FORMAT setting controls body content; frontmatter properties always use wikilink syntax so that wiki-export can reliably parse them.

Only add entries for links added in this cross-linker run — do not touch typed entries that were already present.

Step 5: Score Misc Page Affinity

After the main linking pass, update affinity scores for all pages in misc/ (pages with promotion_status: misc in their frontmatter, or located under the misc/ directory).

For each misc page:

1. Collect outgoing links — all [[wikilinks]] in the page body
2. Collect incoming links — grep the vault for [[misc/<slug>]] and [[<slug>]] references
3. For each linked page (both directions), check if it belongs to a project:
   • Lives under projects/<project-name>/
   • Has a project: frontmatter field matching a project name
4. Group by project name and sum: outgoing_links + incoming_links
5. Update the affinity frontmatter block on the misc page:

affinity:
  obsidian-wiki: 3
  another-project: 1

6. If any project's score ≥ 3: flag this page as a promotion candidate and record it for the report

Efficiency note: only read the full body of misc pages — other pages only need a frontmatter grep to determine their project membership.

Step 6: Report

Present a summary:

## Cross-Link Report

### Links Added: 23 across 12 pages

| Page | Links Added | Confidence | Placement | Relationship Types |
|---|---|---|---|---|
| `projects/my-project/my-project.md` | 3 | EXTRACTED | 2 inline, 1 related | uses ×2, related_to ×1 |
| `entities/jane-doe.md` | 5 | INFERRED | 3 inline, 2 related | extends ×1, uses ×3, related_to ×1 |
| ... | | | | |

### Orphan Pages Remaining: 2
- `references/foo.md` — no incoming or outgoing links found
- `concepts/bar.md` — could not find related pages

### Misc Promotion Candidates: N
Pages in misc/ that have ≥ 3 connections to a single project — ready to be promoted:

| Page | Top Project | Score |
|---|---|---|
| `misc/web-martinfowler-articles-microservices.md` | `obsidian-wiki` | 4 |

To promote: move the page to `projects/<project-name>/references/` and update all backlinks.

### Pages Skipped: 3
- `index.md`, `log.md` — special files
- `_archives/*` — archived content
- `_readouts/*` — derived readouts (wiki-narrate output)

Step 7: Update Log and Hot Cache

One locked call updates the log and the hot cache (the index is unaffected — no pages were created):

obsidian-wiki memory sync CROSS_LINK \
  pages_scanned=<N> links_added=<M> typed_relations_written=<T> \
  pages_modified=<P> orphans_remaining=<Q> \
  misc_affinity_updated=<R> promotion_candidates=<S> \
  --takeaways "Cross-linked 23 mentions across 12 pages; 2 orphans remain."

Never hand-edit index.md, log.md, or hot.md — the command takes the lock that keeps a parallel writer from dropping your update.

See skills/llm-wiki/references/MEMORY.md for the full procedure.

Tips

• Run after every ingest. New pages are almost always poorly connected. This is the fix.
• Be conservative with inline links. Only link the first natural mention, not every occurrence.
• Don't touch pages in _archives/ or _readouts/. Archives are frozen snapshots; readouts are derived output from wiki-narrate, not knowledge pages.
• Respect existing structure. If a page carefully curates its links in a ## Key Concepts section, add to that section rather than creating a separate ## Related.
• Entity pages are link magnets. An entity like jane-doe should be linked from almost every project page. Prioritize these.

QMD Refresh After Vault Writes

QMD is a search index, not the source of truth. If $QMD_WIKI_COLLECTION is empty or unset, skip this step. Run it only after this skill has written or rewritten vault markdown. If QMD refresh fails, do not roll back the vault changes; report the QMD status separately.

Use $QMD_CLI if set; otherwise use qmd.

${QMD_CLI:-qmd} update

If the output says vectors are needed or embeddings may be stale, run:

${QMD_CLI:-qmd} embed

Verify the collection with either:

${QMD_CLI:-qmd} ls "$QMD_WIKI_COLLECTION"

or, when a specific page path is known:

${QMD_CLI:-qmd} get "qmd://$QMD_WIKI_COLLECTION/<page>.md" -l 5

Record one of:

• QMD refreshed: update + embed + verified
• QMD refreshed: update only + verified
• QMD skipped: QMD_WIKI_COLLECTION unset
• QMD skipped: qmd CLI unavailable
• QMD failed: <short error summary>