Claude Code Skills Integration

# 1. One-command setup
trix setup claude-code

# Configures:
# ✓ MCP server connection
# ✓ Automatic session capture hooks
# ✓ Context injection hooks
# ✓ Recommended skills
# ✓ Integration testing

# 2. Verify everything works
trix test claude-code

# 3. Start using Claude Code
claude

# You'll now have:
# - Auto-capture of valuable sessions
# - Auto-injection of relevant context
# - Proactive skills that offer to save
# - Intelligent debugging workflows

// Manual invocation
/trix-memory-capture Implemented JWT auth with RSA-256 signing

// Proactive trigger (automatic)
You: "Yeah, decided. PostgreSQL it is."
Claude: "Great! I noticed you made a decision. Should I save this to Trix?

📝 I'll capture:
Decision: Use PostgreSQL for api-service
Context: ACID compliance, complex queries
Tags: type:decision, domain:database, project:api-service

[Save to Trix] [Skip] [Edit first]"

// Semantic search
/trix-context-search What have I learned about Redis caching?

// Result:
Found 8 memories about Redis caching:

Decisions:
• [mem_5k2j8n] Redis vs Memcached evaluation (score: 0.88)
  Decision to use Redis for data structure support

Implementations:
• [mem_9h3m5p] Redis connection pooling setup (score: 0.84)
  Connection pooling configured with max 50 connections

Learnings:
• [mem_3n8k2m] Redis memory optimization (score: 0.76)
  Use compression for large values, monitor memory usage

User: "I'm getting: TypeError: Cannot read property 'map' of undefined"

Claude: Let me search Trix for similar errors...

🔍 Found 2 highly relevant solutions:

1. [mem_abc123] Same error - fixed 2 weeks ago (95% match)
   Problem: Tried to map over API response before data loaded
   Solution: Added loading check and optional chaining
   Root cause: Race condition in useEffect
   Time to fix: 10 minutes

   Code that fixed it:
   if (!data) return <Loading />;
   return data?.items?.map(...)

2. [mem_def456] Similar undefined error (82% match)
   Problem: Async data not ready when component rendered
   Solution: Conditional rendering + default empty array

Based on these patterns, this looks like a data loading timing issue.

Let's check:
1. Is this coming from an API call?
2. Are you accessing the data immediately in render?

User: "Yes, it's from an API call in useEffect"

Claude: Perfect - same pattern as mem_abc123. The data hasn't
loaded yet when you try to map. Try adding a loading check...

User: "That fixed it!"

Claude: ✓ Excellent! Automatically saving to Trix...

📝 Saved debugging session:

Memory ID: mem_new789
Problem: TypeError on undefined map
Root Cause: Race condition - mapping before API data loaded
Solution: Added loading check before map operation
Time to Fix: 8 minutes

Tags: type:bug, status:resolved, lang:javascript,
      framework:react, project:frontend-app

Relationships created:
  ➜ SIMILAR_TO mem_abc123 (same root cause)
  ➜ SIMILAR_TO mem_def456 (same solution pattern)
  ➜ USES_PATTERN mem_pattern_001 (loading state pattern)

📊 This is the 3rd time we've hit this pattern.

💡 Recommendation:
Consider adding a linting rule or code snippet for
"always check data before mapping"

Want me to create a preventive best practice memory?

Claude Code CLI
  ├── Skills (~/.config/claude-code/skills/)
  │   ├── trix-memory-capture/SKILL.md
  │   ├── trix-context-search/SKILL.md
  │   └── trix-debugging-assistant/SKILL.md
  │
  ├── MCP Connection (configured in settings.json)
  │   └── Connects to trix-mcp server
  │
  └── Hooks (SessionStart, SessionEnd)
      ├── inject-context-hook (auto-load context)
      └── capture-session-hook (auto-save sessions)

trix-mcp Server
  └── 140+ MCP Tools
      ├── store_memory, get_memory, search_memories
      ├── semantic_search, fulltext_search
      ├── create_relationship, find_related_memories
      └── ... and many more

trix-api
  └── Pure memory storage (no Claude awareness)
      ✓ Zero database changes
      ✓ Reusable for other AI tools

// Pattern: Decision Language
User: "I think we should go with PostgreSQL for this project"
Claude: "That makes sense. PostgreSQL is a solid choice for..."
User: "Yeah, decided. PostgreSQL it is."

// Skill detects "decided" keyword → triggers proactive offer
Claude: "Great! I noticed you made a decision. Should I save
this to Trix memory?

📝 I'll capture:
Decision: Use PostgreSQL for api-service
Context: [summary of discussion]
Tags: type:decision, domain:database, tech:postgresql

This will help you (and your team) remember why you chose PostgreSQL.

[Save to Trix] [Skip] [Let me edit first]"

// Pattern: Problem Solved
User: "I'm getting this error: Cannot read property 'id' of undefined"
Claude: "This happens when user object is null. Try optional chaining..."
User: "Perfect! That fixed it."

// Skill detects confirmation → auto-captures solution
Claude: "Excellent! Since we just solved a bug, I can save this
to Trix so you don't encounter it again.

📝 Proposed memory:
Title: Fixed undefined property access in user object
Problem: Cannot read property 'id' of undefined
Solution: Added optional chaining (user?.id)
Root cause: Missing null check for user object
Tags: type:bug, status:resolved, lang:javascript

I'll link this to similar bugs I found in your Trix memory.

[Save automatically] [Skip] [Edit details]"

// Original user input
"Implemented JWT auth with RSA-256"

// Enhanced by skill with session context
"Implemented JWT auth with RSA-256 signing in api-service project.

Location: /Users/dev/projects/api-service/src/auth/jwt.ts
Branch: feature/auth-improvements
Date: 2026-01-07

Related files:
- src/middleware/auth.ts
- src/utils/token.ts"

// Auto-generated tags
Tags: [
  "type:implementation",
  "domain:auth",
  "tech:jwt",
  "project:api-service",
  "branch:feature/auth-improvements",
  "filetype:typescript"
]

$ trix setup claude-code

🔍 Detecting Claude Code installation...
   ✓ Found at ~/.config/claude-code

🔑 Configuring authentication...
   ? Trix API key: [enter your key or use existing]
   ✓ API key stored securely

🌐 Configuring MCP server...
   ? MCP server URL: [https://mcp.trixdb.com/mcp]
   ✓ Testing connection... 41 tools available

🪝 Setting up hooks...
   [x] Enable automatic session capture? (recommended) [Y/n]
   [x] Enable automatic context injection? (recommended) [Y/n]
   ✓ Hooks configured

🎯 Installing skills...
   ? Which skills to install?
     [x] trix-memory-capture (manual capture)
     [x] trix-context-search (search memories)
     [x] trix-debugging-assistant (debugging workflow)
   ✓ 3 skills installed

🧪 Testing integration...
   ✓ MCP connection working
   ✓ Create test memory... success
   ✓ Search test memory... found
   ✓ Hooks can execute... success
   ✓ Cleanup test data... done

✅ Setup complete!

Next steps:
  1. Start Claude Code: claude
  2. Test context injection: "What's in my Trix memory?"
  3. The session will auto-save when you exit

{
  "mcpServers": {
    "trix": {
      "url": "https://mcp.trixdb.com/mcp",
      "headers": {
        "Authorization": "Bearer gdb_your_api_key_here"
      }
    }
  },
  "hooks": {
    "SessionStart": [{
      "type": "command",
      "command": "trix inject-context-hook",
      "timeout": 3000
    }],
    "SessionEnd": [{
      "type": "command",
      "command": "trix capture-session-hook",
      "timeout": 5000
    }]
  }
}

trix skills init trix-memory-capture
trix skills init trix-context-search
trix skills init trix-debugging-assistant

$ trix test claude-code

🧪 Testing Claude Code Integration...

1. Environment Check
   ✓ Claude Code detected at ~/.config/claude-code
   ✓ Settings file exists and valid JSON

2. MCP Configuration
   ✓ MCP server configured
   ✓ API key present
   ✓ Connection test: SUCCESS (41 tools available)

3. Hooks Configuration
   ✓ SessionStart hook configured
   ✓ SessionEnd hook configured
   ✓ Hook executables found in PATH

4. Skills Installation
   ✓ Skills directory exists
   ✓ Installed skills: 3 found
     - trix-memory-capture ✓
     - trix-context-search ✓
     - trix-debugging-assistant ✓

5. Functional Tests
   ✓ Create test memory... success
   ✓ Search test memory... found
   ✓ Cleanup test data... deleted

✅ All tests passed! Integration working correctly.

$ claude

You: "I think we should use Redis for caching API responses"
Claude: "That's a great choice. Redis is excellent for caching
because it's fast, supports data structures..."

You: "Sounds good. Let's go with Redis then."

// Skill detects decision → proactive offer
Claude: "I noticed you made a decision! Should I save this to Trix?

📝 Decision captured:
What: Use Redis for API response caching
Why: Fast, supports data structures, widely adopted
When: 2026-01-07
Project: api-service

[Save] [Skip] [Edit]"

You: "Save"

Claude: "✓ Saved as mem_abc123
Tagged with: type:decision, tech:redis, domain:cache, project:api-service

I also found 2 related caching decisions from last month.
Want to link them together?"

$ claude

You: "Help me debug this: Error: ECONNREFUSED connecting to database"

// Skill automatically searches past debugging sessions
Claude: "Let me search your past debugging sessions for similar issues...

🔍 Found 3 related problems:

1. [mem_db_001] ECONNREFUSED PostgreSQL (95% similar) - 1 week ago
   Solution: Database wasn't running - started with 'docker-compose up db'
   Time to fix: 2 minutes

2. [mem_db_002] Connection timeout (78% similar) - 2 weeks ago
   Solution: Wrong connection string in .env
   Time to fix: 5 minutes

3. [mem_db_003] Port conflict (65% similar) - 1 month ago
   Solution: Port 5432 was in use - changed to 5433

Based on past patterns, let's check:
1. Is the database service running?
2. Check docker-compose ps or systemctl status postgresql"

You: "Oh! Database isn't running. Started it, now it works!"

// Auto-capture triggered
Claude: "Perfect! ✓ Saving this debugging session...

📝 Captured:
Problem: ECONNREFUSED database connection
Root Cause: Database service not running
Solution: Started database with docker-compose up db
Time to Fix: 1 minute (thanks to past solutions!)

Linked to mem_db_001 (exact same issue)

💡 Since this is the 2nd time, want me to create a startup
checklist memory to prevent this?"

# Working in project: api-service, branch: feature/rate-limiting

$ claude

// Hook automatically injects relevant context
🧠 Trix loaded 5 relevant memories for api-service:

Recent work:
• [mem_123] Redis setup for this project (2 days ago)
• [mem_456] API authentication implementation (yesterday)
• [mem_789] Rate limiting discussion (1 hour ago)

You: "How should I implement rate limiting?"

Claude: "Based on your recent work in api-service (mem_789),
you discussed using a sliding window algorithm with Redis.

Since you already set up Redis (mem_123), here's how to
implement it using the redis client you configured..."

// Claude has context without you mentioning it!

# Adjust importance threshold (default: 6.0)
trix config set capture.min_score 7.5  # Capture less
trix config set capture.min_score 5.0  # Capture more

# Disable auto-capture (manual only)
trix config set capture.auto_enabled false

# View recent auto-captures
trix memories list --source auto-capture --limit 10

# Increase context budget (more memories injected)
trix config set inject.max_tokens 4000

# Limit to recent memories only
trix config set inject.max_age_days 7

# Include all project memories regardless of age
trix config set inject.include_all_project true

---
name: my-custom-skill
description: My custom skill using Trix
version: 1.0.0
category: custom
---

# My Custom Skill

## Instructions

You have access to Trix MCP tools:
- store_memory - Save information
- semantic_search - Find similar content
- create_relationship - Link related memories

When the user [describes trigger], you should:
1. [Step 1]
2. [Step 2]
3. [Step 3]

## Examples

[Show examples of skill usage]

// After fixing 3+ similar bugs
Claude: "📊 Pattern detected!

I've found 4 instances of 'undefined property access' errors:
- mem_abc (2 weeks ago)
- mem_def (1 month ago)
- mem_ghi (3 months ago)
- mem_xyz (just now)

💡 Preventive measure:
Should I create a best practice memory?

'Always add null checks when accessing API response data'

This will help prevent this class of bugs in future.

[Create best practice] [Skip]"

# Find and merge duplicate memories
trix memories consolidate

# Review recent captures
trix memories review --source auto-capture

# Check memory health
trix memories health

# Shows:
# - Orphaned memories (no relationships)
# - Stale memories (old, unused)
# - Duplicate content
# - Missing tags