Wiki Lint

Purpose

Audit the wiki for structural and quality issues. Catches contradictions, stale content, orphan pages, duplicate concepts, missing sections, and broken [[wikilinks]] BEFORE downstream agents act on stale or conflicting data.

Run weekly (default Mondays via scheduled task). Outputs lint-report.md for human review — never auto-fixes.

When To Invoke

• Weekly schedule (default: Mondays at 7am via /mxm-tasks)
• User asks for "wiki health check" or "audit the wiki"
• Before any major content release
• After bulk ingest of >50 pages

Checks Performed

1. Contradictions

For every entity or concept appearing in multiple pages, compare claims using:

• mempalace_kg_query(entity, direction="both") — find all triples for that entity
• mempalace_search(entity) — find all pages mentioning it
• Flag pages making contradictory factual claims

2. Stale Content

• Pages where updated date > 90 days old AND confidence: high
• These are high-confidence claims that may have decayed
• Output: list with last-updated date, suggest re-verification

3. Orphan Pages

• Pages with zero inbound [[wikilinks]] from other pages
• Use mempalace_find_tunnels and graph queries
• Suggest either deletion or linking from related pages

4. Duplicate Concepts

• Different page names covering the same idea (semantic similarity)
• Use mempalace_check_duplicate(content, threshold=0.85)
• Suggest merge with target page

5. Missing Mandatory Sections

For every page, verify presence of:

• ## TLDR
• ## Counter-Arguments
• ## Related (with at least 2 wikilinks)
• ## Sources

Pages missing any: flag for completion.

6. Broken Wikilinks

• [[wikilinks]] pointing to non-existent pages
• Suggest: create stub page, fix link target, or remove

7. Confidence Inconsistency

• Pages with confidence: high but explored: false — invalid combination
• Suggest: lower confidence to medium OR run /mxm-wiki explore to verify

8. AI-Tell Contamination

• Any page containing phrases from .brand-foundation/personal/ai-tells.md
• Suggest: rewrite that paragraph

Output Format

Wiki Lint Report — [YYYY-MM-DD HH:MM]
Total pages audited: [N]
Issues found: [N]

## Contradictions ([N])
- entity: <X>
  - page: [[A]] claims: "<claim 1>"
  - page: [[B]] claims: "<claim 2 — contradicts>"
  - resolution: HUMAN_REVIEW

## Stale Content ([N], confidence:high + updated >90d)
- [[<page>]] (updated YYYY-MM-DD, [N] days old) — re-verify

## Orphan Pages ([N])
- [[<page>]] — 0 inbound links — suggest delete OR link from [[parent-concept]]

## Duplicate Concepts ([N])
- [[<page-A>]] ≈ [[<page-B>]] (similarity 0.92) — suggest merge into [[<page-A>]]

## Missing Sections ([N])
- [[<page>]] — missing: Counter-Arguments, Sources

## Broken Wikilinks ([N])
- [[<page>]] → [[<missing-target>]] — suggest: create stub OR fix typo

## Confidence Inconsistency ([N])
- [[<page>]] — confidence:high + explored:false → set medium OR run /mxm-wiki explore

## AI-Tell Contamination ([N])
- [[<page>]] paragraph 3 contains banned phrase: "delve into the landscape"

---
Confidence: 🟢 HIGH (full corpus scanned, all checks passed) | 🟡 MEDIUM (sample only, time-bounded) | 🔴 LOW (corpus too large to fully audit; partial report)
Next: human review at lint-report.md, then `/mxm-wiki explore` for any flagged unverified pages.

Output is written to lint-report.md in the project root (under .mxm-knowledge/ if using project-scoped knowledge).

Critical Rule

wiki-lint NEVER auto-fixes. All changes require human review. The skill produces findings; the operator decides resolution.

MemPalace Integration

mempalace_kg_query(...)        # contradictions
mempalace_check_duplicate(...) # duplicates
mempalace_find_tunnels(...)    # orphan / linking analysis
mempalace_search(...)          # full-corpus audit pass
mempalace_kg_timeline(...)     # staleness via update history

Handoff

• Contradictions found → escalate to wiki-explore (human verification queue)
• Stale high-confidence pages → notify wiki-ingest to re-check sources
• Compliance-sensitive contradictions → loop compliance-officer
• AI-Tell contamination → notify brand-guardian for rewrite

Maxim Behavioral Layer

• Confidence tagging: 🟢 HIGH on full-corpus audit | 🟡 MEDIUM on sampled | 🔴 LOW on partial
• Frameworks: LLM Wikid quality control pattern
• Anti-noise: thresholds tuned to minimize false positives (similarity 0.85 for duplicates, 90d for staleness)

Source References

• FINAL_RELEASE_CHANGES/Brandvoice-wiki-example.md — wiki-lint pattern
• documents/reference/AGENTS.md — wiki trust levels
• .brand-foundation/personal/ai-tells.md — AI-tell scanner source

---

Copyright (c) 2026 iSystematic Inc. Maxim product. BSL 1.1.