Quick Start

Learn the core Kyew workflow in 5 minutes: store memories, detect patterns, generate skills, and get context-aware suggestions.

The Learning Loop

Kyew implements a simple but powerful learning loop:

Experience → Memory → Pattern → Skill → Application

1. Experience: You work on tasks and solve problems
2. Memory: Store observations about what worked
3. Pattern: Detect repeated approaches across memories
4. Skill: Codify patterns into reusable knowledge
5. Application: Skills guide future similar work

Step 1: Store Memories

As you work, store observations about your decisions and outcomes:

"remember that I debugged CORS by adding Mcp-Session-Id to Access-Control-Expose-Headers"

Claude uses the remember tool with context:

{
  "observation": "Debugged CORS by adding Mcp-Session-Id to Access-Control-Expose-Headers",
  "domain": "cloudflare",
  "task_type": "debugging",
  "outcome": "success"
}

What to Remember

Good memories are specific and actionable:

• ✅ "Fixed authentication by checking the JWT expiry in middleware"
• ✅ "Deployment failed because wrangler.toml was missing the containers config"
• ✅ "API tests need to mock the database connection in vitest.setup.ts"

Avoid vague or obvious statements:

• ❌ "The code worked"
• ❌ "I wrote some tests"

Step 2: Recall Memories

Later, retrieve relevant memories when facing similar situations:

"recall memories about CORS debugging"

Claude uses semantic search to find related memories:

{
  "memories": [
    {
      "observation": "CORS issue was fixed by adding Access-Control-Expose-Headers",
      "domain": "cloudflare",
      "outcome": "success",
      "created": "2024-01-15T10:30:00Z"
    }
  ]
}

Step 3: Detect Patterns

After accumulating memories, analyze them for patterns:

"analyze patterns in the cloudflare domain"

The system finds repeated approaches:

{
  "patterns": [
    {
      "description": "CORS debugging pattern",
      "memory_count": 3,
      "confidence": 0.85,
      "memory_ids": ["mem-1", "mem-2", "mem-3"]
    }
  ]
}

Step 4: Generate Skills

Convert a detected pattern into a reusable skill:

"generate skill from those CORS debugging memories"

The system creates a draft skill:

{
  "skill": {
    "name": "Cloudflare CORS Debugging",
    "status": "draft",
    "instructions": "When debugging CORS issues on Cloudflare Workers:\n1. Check Access-Control-Allow-Origin header\n2. Verify Access-Control-Expose-Headers includes custom headers\n3. Ensure preflight OPTIONS requests are handled..."
  }
}

Step 5: Approve Skills

Review and approve skills to activate them:

"show pending skills"

Then approve:

"approve the CORS debugging skill"

Once approved, the skill becomes active and will be suggested in relevant contexts.

Step 6: Get Contextual Help

When working on similar tasks, request context-aware guidance:

"get context for debugging API issues on Cloudflare"

Claude retrieves relevant memories and suggests applicable skills:

{
  "memories": [...],
  "skills": [
    {
      "name": "Cloudflare CORS Debugging",
      "relevance": 0.92,
      "instructions": "..."
    }
  ]
}

Complete Example

Here's a full workflow session:

# Day 1: Working on a project
User: remember that the Ghost CMS API requires the Admin API key in the Authorization header as Ghost {{api_key}}

# Day 3: Another API integration
User: remember that Ghost webhooks need to be verified using the webhook secret with HMAC-SHA256

# Day 7: More Ghost work
User: remember that Ghost content API responses are paginated with meta.pagination

# Day 10: Analyze patterns
User: analyze patterns in the ghost domain

# Result: Pattern detected with 3 memories, 0.9 confidence

# Generate a skill
User: generate skill from those Ghost API memories

# Review and approve
User: show the pending Ghost API skill
User: approve it

# Future use
User: I need to integrate with Ghost API
User: get context for Ghost API integration
# → Returns the approved skill with step-by-step guidance

Tips for Effective Learning

Be Specific

More detail = better pattern matching:

# Good
remember that vitest tests need isolatedModules: true in tsconfig for ESM support

# Better
remember that vitest tests fail with "Cannot use import statement outside a module" and the fix is setting isolatedModules: true in tsconfig.json

Include Outcomes

Always specify whether something worked:

remember that using $map instead of .map() fixed the JSONata expression - outcome success

Use Consistent Domains

Group related work under consistent domain names:

remember ... domain "cloudflare-workers"
remember ... domain "cloudflare-workers"  # Same domain
remember ... domain "cloudflare"          # Different domain - might miss patterns

Review Before Approving

Always review generated skills before approval. The AI synthesis is helpful but may miss nuances only you understand.