Import skill: deduplicating-vault (Obsidian vault deduplication)

[randlee]

Skill Import Request

Skill name: deduplicating-vault
Agent: vault-duplicate-scan
Script: find_vault_duplicates.py

Local path (source)

.claude/skills/deduplicating-vault/
├── SKILL.md
└── scripts/
    ├── find_vault_duplicates.py
    └── tests/
        └── test_find_vault_duplicates.py

.claude/agents/vault-duplicate-scan.md
.claude/agents/registry.yaml

Full path on disk: /Users/randlee/Documents/one/.claude/skills/deduplicating-vault/

What it does

Scans an Obsidian vault for:

1. Failed clips — Obsidian Clipper stubs with title but no body, matched against a real copy via normalized title similarity
2. Exact duplicates — identical body content (SHA-256 hash), different frontmatter
3. Near-duplicates — body similarity ≥ 80% (configurable) via difflib.SequenceMatcher

Reports findings grouped by type with per-item recommendations:

• Which file to keep (prefers organized folder > Clippings/, richer tags, older created date, longer body)
• Tags to merge from deleted copy into keeper
• Folder placement recommendation (CORRECT_LOCATION / MOVE_RECOMMENDED / UNCERTAIN_LOCATION)

User approves actions; no auto-delete.

Interface

/deduplicating-vault                        # full scan, defaults
/deduplicating-vault --threshold=0.9        # stricter match
/deduplicating-vault --folder=Claude/       # scope to subfolder
/deduplicating-vault --format=json          # raw JSON output

Architecture

• No external dependencies — Python stdlib only (difflib, hashlib, pathlib, json, argparse, re)
• Pure functions — all core logic independently unit testable; I/O only at load_vault_files()
• Session-scoped hash index — SHA-256 index built in-memory at startup, discarded on exit
• quick_ratio() pre-filter — eliminates low-similarity pairs before full SequenceMatcher.ratio() (~3× speedup)
• Fenced JSON agent output — conforms to claude-code-skills-agents-guidelines v0.7
• 27 unit tests — all pure functions tested with in-memory strings, no vault access

Guidelines conformance

Architecture follows: synaptic-canvas/docs/claude-code-skills-agents-guidelines.md (v0.7)

Convention	Value
Skill name (gerund)	deduplicating-vault
Agent name (noun-verb)	vault-duplicate-scan
Agent output	fenced JSON
Registry entry	registry.yaml (included)
External deps	none
Unit tests	27 (all pass)

Request

Import this skill into the Synaptic Canvas Obsidian plugin marketplace so it can be discovered and installed via gh skill install or equivalent marketplace tooling.