wiki-inline-link-audit

name	wiki-inline-link-audit
description	USE THIS SKILL whenever the user wants to: add wikilinks to unlinked mentions, improve graph connectivity, clean up Related sections, or cross-reference pages in a markdown vault. Triggers on 'link mentions', 'inline links', 'connect pages', or 'wiki linking'.

Wiki Inline Link Audit

Purpose

Find terms in wiki page body text that reference other wiki pages but aren't wikilinked. Goal: Wikipedia-style inline linking where entities are naturally linked when mentioned in flowing text — not backlink sections at the bottom.

When to Use

• After bulk imports of new wiki pages (people, projects, places)
• Periodically as wiki grows and new cross-references emerge
• When user wants to improve graph connectivity

Steps

1. Build Term Index

# For each wiki page, collect: slug -> {title, aliases, kind}
# Build term_to_slug: lowercase(term) -> slug for all titles + aliases
# Exclude course pages from auto-link targets if user has course indexes

Course pages exclusion: If the wiki has course-type pages (slugs matching faculty-course-number patterns like asia-101, cpsc-110), exclude them from auto-link targets. User likely wants faculty-level index pages instead of linking to individual courses. See "Course Index Pages" below.

Process longest terms first: Sort term_to_slug by term length descending. This prevents shorter substrings from being linked before longer complete phrases (e.g., "Jane Smith" should match before "Jane").

2. Scan Body Text

For each page, extract body (strip frontmatter ---...---) and find paragraphs of flowing text. Exclude:

• Lines starting with #, |, - , * , >,
• Numbered list lines (1.)
• Content inside existing [[wikilinks]] (check unclosed [[ before match position)
• Heading lines
• ## Sources section content

3. Match Terms

• Chinese terms: exact character match (no word boundaries)
• English terms: \b word boundary match, case-insensitive
• Skip: own page (don't self-link), terms < 3 chars, generic/common terms

Generic terms to exclude (too vague to link):

history, professional, academic, education, student, production, media, digital,
community, design, content, analysis, technology, science, management, development,
research, network, project, study, learning, writing, creative, art, film, video,
culture, language, introduction, foundation, modern, contemporary, critical, early,
global, information, society, family, environmental, geography, music, economics,
politics, singing, photography, cinema, horror, fiction, visualization, policy,
meteorology, anatomy, physiology, chemistry, program, interactive, storytelling,
audiences, industries, religions, thought, approaches, career, applied, systematic,
methods, networks, crowds, communities, arts, visual, second language, english,
new media, human, cap, seminar

4. Categorize Results

• Batch-fix patterns: Large groups of pages with identical unlinked terms (e.g., 100+ person pages all missing [[wenzhou-no22-high-school]]). Safe to auto-fix.
• High-value inline links: Key pages (projects, places, prominent people) with specific unlinked concepts. Review before fixing.
• Exclusions: Course pages linking to each other may need index pages instead. Skip auto-linking course-to-course.

5. Apply Fixes

Replace term with [[target-slug|term]] in body text. Use |display form to preserve original text.

Precautions:

• Process matches from end to start (preserves positions)
• Critical — prevent split links: When a longer match exists (e.g., "Jane Smith" matches jane-smith), skip shorter substrings ("Jane", "Smith") at the same position. After batch fix, scan for [[slug|X]] [[slug|Y]] patterns and merge to [[slug|X Y]].
• Deduplicate: same (target, term) per page only once
• Chinese text before [[ without space is correct (中文不需要空格)
• Same link appearing twice in one paragraph is fine if different context

Pitfalls

• Split links (critical): If "Jane" and "Smith" are both indexed as terms for jane-smith, the script may link them separately creating [[jane-smith|Jane]] [[jane-smith|Smith]]. Fix: when processing, skip shorter terms that are substrings of already-matched longer terms at the same position. Always run a post-fix scan: r'\[\[([^|]+)\|([^]]+)\]\]\s*\[\[\1\|([^]]+)\]\]' and merge to [[slug|X Y]]. In practice this appeared in 4 files after a 189-file batch fix.
• Student roster space bug: When programmatically inserting [[slug|display]] into table cells, trailing spaces end up inside brackets ([[slug |display]] or [[slug|display ]]). These render correctly in Obsidian but are broken links in the graph. After batch linking tables, always scan and fix: re.sub(r'\[\[([^\]|]+?)\s+(\|[^\]]+?)?\]\]', lambda m: f'[[{m.group(1)}{m.group(2) or ""}]]', content). In one case this caused 109 broken links from a single roster file.
• Surname columns in tables: When linking names in rosters/databases, link only name columns (中文名, English). Do NOT link 姓/名 (surname/given name) columns — common surnames like Yang, Wu, Chen will match the wrong person. After initial linking, unlink these columns: re.sub(r'\[\[([^\]|]+\|)?([^\]]+)\]\]', r'\2', cell).
• Generic terms: "history", "film", "media" etc. will match tons of false positives. Always maintain and expand the exclusion set.
• Course pages: Don't auto-link course names to individual courses. Create faculty/institution index pages and link there instead.
• Table content: Skip table rows (| lines) — tables are structured data, not flowing prose.
• Aliases are broad: Some aliases are very short or common. Filter carefully.
• Sandbox privacy masking: When running in sandboxed environments, phone numbers and other PII may be displayed with **** masking. The actual file content may be correct — verify with xxd or hex() before assuming data is corrupted. The sandbox masks DISPLAY only, not file content.

Tag Standardization

After wiki grows, tag sprawl is common (30+ unique tags). Standardize to a two-layer system.

Layer Tags (from kind)

Map directly to page type: person | project | place | topic | concept | map

Domain Tags (optional, 0-2 per page)

Broad domains relevant to the wiki owner: e.g., ai | media | startup | transit | writing

Tags to Remove or Merge

• Tech stack tags (tauri, rust, nextjs, react-native, supabase) — belong in article body
• Redundant name tags (owonetwork, citang) — page title already identifies the entity
• One-off tags (maintenance, index, social-circle) — not useful for filtering
• Merge close variants: school → school, media-studies → media, token-tracking → ai
• Fix quote inconsistencies: 'person' → person

Implementation

Build tag map (old → new), apply via regex on frontmatter tags: line. Document the taxonomy in system/wiki/schema.md so future agents follow it.

Stub Page Index

When many pages are stubs (< 10 body lines), maintain a topics/stub-index.md listing them with title, kind, and line count. Group by kind. This serves as a task list for future enrichment and cross-reference point when new inbox imports arrive.

Schema Persistence

Conventions discovered during wiki work MUST be written into system/wiki/schema.md and system/wiki/workflows.md. Other agents and future sessions won't know about session-only decisions. If you discover a new convention (e.g., "no ## Related sections"), immediately add it to schema.md. Otherwise the next agent will reintroduce the pattern you just removed.

Removing ## Related Sections

When the wiki has ## Related sections at the bottom of pages (a list of wikilinks), these should be removed once inline links are in place — Wikipedia doesn't have a "related pages" section; links are woven into body text.

Strategy

1. Categorize Related sections by pattern before removing:

   • Count how many links in Related are already covered in body text
   • Pages where ALL Related links are in body → safe to batch-remove
   • Pages with uncovered links → need manual integration first
   • Common patterns: person-page-only (45 pages), two-links (123 pages), multi-links (15 pages)

2. Batch removal: For pages where all Related links are already covered in body, remove ## Related\n...\n block directly. Clean up triple newlines.

3. Manual integration: For uncovered links, add them naturally to body text before removing the section.

4. Verify: After removal, confirm zero ## Related sections remain.

Encyclopedic Wiki Review Checklist

When reviewing a wiki from an encyclopedia perspective:

1. Inline linking: Are entities linked when naturally mentioned in body text?
2. No backlink sections: ## Related sections should not exist; links should be inline.
3. Orphan pages: Pages with zero inbound links (excluding maps/hubs). For person pages, check if they should link to each other or be linked from hub pages.
4. Broken links: [[links]] pointing to non-existent pages. Check for trailing spaces inside brackets ([[slug ]]).
5. Stub pages: Pages with < 10 body lines. Maintain a stub-index.md for tracking.
6. Bloated pages: Pages > 150 lines that should be split (by topic, not by time period).
7. Isolated pages: Pages with zero inbound AND zero outbound links — enrich and link from related pages.
8. Missing Sources: Core pages lacking ## Sources section. Course index pages are exempt.
9. Tag distribution: Tags should have a defined taxonomy (see schema.md). One-off tags and tech stack tags should be removed.
10. Date-based headings: Sections named by date (## 2024-01) indicate chronological structure instead of thematic. Rewrite.
11. WeChat nicknames in aliases: Scan for emoji in aliases, parenthetical nicknames in H1, known nickname patterns.
12. Inbox references in Sources: No inbox/ paths should appear in ## Sources sections.

Collapsible Callouts for Long Data

When a page has long data tables (email lists, phone numbers, credentials), wrap them in Obsidian collapsible callouts instead of leaving them as open tables:

> [!info]- All emails (21)
> | Address | Purpose |
> |---------|---------|
> | ...     | ...     |

Duplicate Page Merging

When the same entity exists as multiple pages (e.g., projects/bc-ai-ecosystem.md AND topics/bc-ai-ecosystem.md):

1. Identify which version is richer (more lines, more content)
2. Merge unique content from the other version (aliases, sources, descriptions)
3. Keep the richer version, update its kind and tags if needed
4. Delete the duplicate file
5. Verify: no broken links (slug is the same, so existing links still resolve)

WeChat Nickname Convention

Social media nicknames (微信昵称, Instagram handles, etc.) must NOT go in:

• aliases: in frontmatter
• H1 page title (no parentheses with nickname)

Nicknames belong in body text under Contact or personal info section:

# wang-wu

某中学同学。社交昵称"Dreamer"。

NOT:

# wang-wu (社交昵称)   ← WRONG
aliases: ['wang-wu', '社交昵称']  ← WRONG

Aliases are for real name variants and common transliterations only (e.g., "James" for James Doe, "Alex" for Alex Lee).

When auditing person pages, scan for:

• Parentheses in H1 containing non-real-name text
• Aliases containing emoji (🐼, 🤫, 🦠, 💕)
• Aliases matching known nickname patterns (cute phrases, handles, WeChat-style names)

Inbox Reference Rule

## Sources sections must NEVER reference inbox/ paths. Inbox is a temporary staging area — all content is deleted after distillation into sources/ and wiki/.

When importing data, copy raw files to sources/ first, then reference sources/ paths in wiki Sources sections. Common migrations:

• inbox/identity/contacts.vcf → sources/identity/contacts.vcf
• inbox/Whatsapp/*.txt → sources/chats/*.txt
• inbox/facebook-* → already absorbed into wiki, remove reference

Course Index Pages

When the wiki has many course-type pages, don't link to individual courses from other pages. Instead create a hierarchy:

courses.md (总索引)
├── courses-university.md (university courses, grouped by Faculty)
│   ├── ASIA (9 courses)
│   ├── CHIN (4 courses)
│   └── ...
└── courses-school.md (school courses, grouped by type)
    ├── BC 省课程
    └── 中方课程

Each course page links to its faculty/institution index in ## 相关链接. Other pages (people, projects) link to the index page, not individual courses.

Flighty CSV Import

When importing Flighty flight export CSV:

1. Save raw CSV to sources/flights/ with date slug
2. Parse CSV: extract Date, Airline, Flight number, From, To, Aircraft, Seat, Notes
3. Create a source-layer summary page with a flight table and airport code to city mapping
4. Create/update a wiki/topics/transportation.md page with stats (total flights, common routes, cities visited)
5. Add/update the Travel section on the person page (some-person.md) with city list linking to existing place pages
6. Register new pages in system/wiki/index.md

Airport codes should link to place pages where they exist (e.g., [[vancouver|Vancouver]] for YVR). Cities without place pages remain plain text.

Schema Write Discipline

When you discover a new convention during wiki work (e.g., "no ## Related sections", "WeChat nicknames not in aliases", "no inbox/ in Sources"), write it into system/wiki/schema.md or system/wiki/workflows.md immediately, in the same session. Do not defer. Other agents and future sessions have no memory of session-only decisions. A convention not written down will be violated by the next agent.