AgentSkillsCN

Vault Search

Vault Search

SKILL.md

Vault Search Skill

Architecture: 3 Layers

Layer 1: Direct ripgrep (80% of queries)

bash
# Fast, direct patterns
rg "3-2-1 technique" --glob '*.md'              # content search
rg "tags:.*psychology" --glob '*.md'            # tag search
rg "created: 2026-01" --glob '*.md'             # date search
rg "^- \[ \].*📅 2026-01-27" --glob '*.md'      # tasks due today

Layer 2: Python script (date math & language detection)

python
# Only for what regex can't do:
"recent week" → datetime.now() - timedelta(7)
"overdue" → filter(date < today)
"找3-2-1技巧" → Chinese term extraction

Layer 3: AI agent (orchestration)

  • •Understands intent
  • •Chooses tool (ripgrep vs Python)
  • •Combines results
  • •Formats output

Direct Patterns (Use ripgrep)

Query TypePattern
Contentkeyword or (word1|word2)
Tagstags:.*tagname
Date (YYYY-MM-DD)created: 2026-01-27
Date rangecreated: 2026-01
Tasks^- \[ \]
Tasks due today^- \[ \].*📅 2026-01-27
High priority^- \[ \].*⏫

When to Use Python Script

NeedWhy Python
"recent week"Date calculation
"overdue tasks"Date comparison
Language-specificPattern variants
Fallback datesFile system mtime

Language-Aware Term Extraction

  • •Chinese queries → extract contiguous CJK terms for rg
  • •English queries → extract alnum tokens (3+ chars)
  • •Synonym expansion is handled by the agent, not this script

Date Field Patterns

Most common in Clippings:

  • •created: 2026-01-27 (YYYY-MM-DD)

Fallback:

  • •File system modified time if no frontmatter

Task Emojis

  • •📅 Due date
  • •🛫 Start date
  • •⏳ Scheduled
  • •⏫ High priority
  • •🔼 Medium
  • •🔽 Low

CLI Usage

bash
# Notes
vault_search.py "recent week psychology"
vault_search.py "找3-2-1技巧"

# Tasks
vault_search.py "tasks due today"
vault_search.py "overdue tasks"

# Raw output for piping
vault_search.py "3-2-1" --raw | xargs -I {} echo "{}"