obsidian-semantic-lint

name	obsidian-semantic-lint
description	Performs high-level semantic health checks on the vault, identifying potential contradictions, stale summaries, and structural anomalies in the knowledge graph.
version	1.0.0
author	Apollo
license	MIT
metadata	{"hermes":{"tags":["obsidian","lint","hygiene","knowledge-graph","semantics"],"related_skills":["obsidian","obsidian-hygiene"]}}

obsidian-semantic-lint

Overview

This skill is responsible for running Tier-3 (semantic) health checks on the Obsidian vault. Unlike obsidian-hygiene, which focuses on structural issues like broken links and file locations, this skill analyzes the content of the notes to find potential issues in the knowledge graph.

This skill absorbs the Semantic Lint workflow from the deprecated llm-wiki skill.

When to Use

• On-demand: When Justin asks to "run a wiki lint" or "check for contradictions". It should always be run after obsidian-hygiene to ensure a clean structural baseline.
• Via Cron: A monthly cron job runs this skill to generate a report of potential issues for review.

Checks Performed

This skill runs the wiki_semantic_lint.py script (which completely excludes the Inbox/ and inbox/ folders to let Justin triage those manually), checking for several classes of semantic problems:

1. Deterministic Orphans: Maturity orphans (notes under Notebook/ in mature categories) are only flagged if they have neither inbound nor outbound links (completely isolated). Notes that have outbound links but no inbound links are NOT counted as orphans.
2. Stale Summaries: Source notes whose underlying Reading notes have been updated more recently than the Source summary itself.
3. Promotion Gaps: High-level Beliefs that are not supported by a chain of Thoughts or Concepts, indicating a potential leap in logic.
4. Circular Reasoning: Identifies link chains that form a closed loop (e.g., A → B → C → A), which can be a sign of a tautological argument.
5. Contradiction Candidates: Flags pairs of notes (especially Beliefs and Thoughts) that make opposing claims. This is a heuristic and requires human review.

Workflow

1. Baseline with Structural Lint: When run on-demand, first ensure obsidian-hygiene has been run to fix any structural problems.
2. Execute Script: Run the wiki_semantic_lint.py script.
3. Generate Report: The script outputs a report in markdown format to Utilities/reports/semantic-lint-YYYY-MM-DD.md.
4. State Management: The script also updates a state file at ~/.hermes/state/semantic_lint_last.json to track findings between runs.
5. Review (Agent/User): The agent should review the "Contradiction Candidates" section of the report. It should never auto-edit notes based on these findings, especially not Beliefs. The findings should be presented to Justin for review.

Pitfalls

• Auto-Editing: Never automatically change the content of a note based on a semantic lint report. These are candidates for review, not definitive errors.
• Running Before Hygiene: Running semantic checks on a structurally unsound vault can lead to confusing or inaccurate results. Fix the simple problems first.
• Confusing with Structural Lint: Remember the division of labor: obsidian-hygiene is for structure (files, links, IDs), obsidian-semantic-lint is for content and meaning.
• Scanning the Inbox: The semantic lint scan (as part of EEIRP/EIIRP) must always ignore the Inbox/ and inbox/ directories. These notes are un-triaged and temporary, and scanning them produces premature or false-positive issues.