tools-only Context Management Skill
Purpose: Seamless integration between your Obsidian knowledge base and Claude Code sessions. Find relevant notes, load project context, and manage captures from Telegram.
Quick Reference
obs CLI Commands
# Search
obs search "query" # Full-text search
obs search --tag "project/pai" # Tag-based search
obs semantic "concept" # Semantic similarity search
# Read/Write
obs read "Note Title" # Read a specific note
obs write "Title" --tag inbox # Create a new note
# Browse
obs tags # List all tags
obs tags --counts # Tags with usage counts
obs recent # Recently modified notes
obs incoming # Unprocessed inbox items
# Context
obs context project-name # Load project context
# Embeddings
obs embed # Build/update embeddings
obs stats # Embedding statistics
ingest CLI Commands
# Telegram capture
ingest poll # Fetch new messages
ingest process # Process pending → notes
ingest status # View message states
# Maintenance
ingest retry # Retry failed messages
ingest clear # Clear all (testing)
Workflows
1. load-context.md
Purpose: Load project-specific context into current session Trigger: "load context for X", "get project context", "what do I know about X"
2. search-notes.md
Purpose: Search personal notes using text, tags, or semantic similarity Trigger: "search my notes for", "find notes about", "semantic search"
3. process-incoming.md
Purpose: Review and process captured content from Telegram Trigger: "check incoming", "process captures", "what's in my inbox"
4. capture-content.md
Purpose: Capture new content to Obsidian via direct write Trigger: "save this to obsidian", "capture this", "add to my notes"
Routing Logic
User Intent → Workflow Selection
"Load context for PAI project" → load-context.md
"Search notes about embeddings" → search-notes.md
"What's in my inbox?" → process-incoming.md
"Save this to my notes" → capture-content.md
Integration with Claude Code
This skill enables context-aware coding sessions:
- Start of session: Load relevant project context
- During coding: Search notes for reference material
- After research: Capture findings to Obsidian
- End of session: Auto-saved via hooks
Example Session Flow
User: "Load context for the API redesign project"
→ obs context api-redesign
→ Returns: Related notes, decisions, meeting notes
User: "What did we decide about authentication?"
→ obs semantic "authentication decision"
→ Returns: Relevant note excerpts
User: "Save this architecture decision to my notes"
→ obs write "API Auth Decision" --tag "project/api" --tag "decision"
→ Creates note in vault
Vault Structure
Expected Obsidian vault organization:
~/Documents/personal/
├── Inbox/ # New captures land here
├── Projects/ # Project-specific notes
│ └── project-name/
├── Reference/ # Technical references
├── Archive/ # Completed/old items
└── attachments/ # Images, PDFs, etc.
Tag Conventions
| Tag Pattern | Purpose |
|---|---|
project/<name> |
Project association |
incoming |
Unprocessed captures |
decision |
Architectural decisions |
meeting |
Meeting notes |
reference |
Reference material |
todo |
Action items |
Configuration
Environment variables (in ~/.claude/.env):
OBSIDIAN_VAULT_PATH=~/Documents/personal
OPENAI_API_KEY=sk-... # For embeddings & vision
State databases:
- Embeddings:
~/.claude/embeddings.db - Ingest state:
~/.claude/ingest-state.db
Troubleshooting
Semantic search not working
obs stats # Check if embeddings exist
obs embed # Rebuild embeddings
Telegram captures not appearing
ingest test # Verify bot connection
ingest poll # Fetch new messages
ingest status # Check message states
Note not found
obs tags --counts # See what tags exist
obs recent # Check recent files
Related Skills
- research - Multi-source web research
- fabric - Pattern-based content processing
References
- Obsidian: https://obsidian.md
- OpenAI Embeddings: https://platform.openai.com/docs/guides/embeddings