Get Pattern

Get Pattern - Retrieve Application Knowledge

What This Skill Does

Retrieves established application patterns (architecture, procedures, conventions) for the Neural Data Platform using AgentDB's semantic skill search.

Use this BEFORE implementing anything to ensure you follow project standards.

Quick Reference

# Search for patterns by description
npx agentdb skill search "domain adapter pattern" 5

# Fallback: search reflexion episodes for past experiences
npx agentdb reflexion retrieve "how to add a stream" --k 5 --only-successes

# View all stored patterns
npx agentdb db stats

Primary Method: Skill Search

npx agentdb skill search "<query>" <k>

Parameters

Examples

# Find architecture patterns
npx agentdb skill search "domain adapter pattern" 5

# Find deployment procedures
npx agentdb skill search "deploy to raspberry pi" 3

# Find naming conventions
npx agentdb skill search "naming conventions streams fields" 3

# Find troubleshooting guides
npx agentdb skill search "mqtt data not appearing" 5

Fallback Method: Reflexion Retrieve

If no skill patterns exist, search past experiences:

npx agentdb reflexion retrieve "<query>" --k 5 --only-successes --synthesize-context

Parameters

Examples

# Find successful similar work
npx agentdb reflexion retrieve "HTTP source implementation" \
  --k 5 \
  --only-successes \
  --min-reward 0.7

# Get synthesized context
npx agentdb reflexion retrieve "timescaledb schema" \
  --k 10 \
  --synthesize-context

Pattern Categories

Interpreting Results

Results from skill search include:

High-value patterns: Success Rate > 80% AND Uses > 3

Typical Workflow

# 1. Search for existing patterns
npx agentdb skill search "what I'm about to implement" 5

# 2. If found: Follow the pattern
# 3. If not found: Check reflexion for past experiences
npx agentdb reflexion retrieve "similar task" --k 5 --only-successes

# 4. After work: Record feedback
npx agentdb reflexion store "feature-id" "task" 0.9 true "Pattern worked well"

# 5. If you discovered something new: Save it
npx agentdb skill create "pattern-name" "description" "optional-details"

CRITICAL: Record Pattern Usage

After using a pattern, always use the reflexion skill to record whether it helped:

# Pattern worked well
npx agentdb reflexion store "dp-004" \
  "Used domain-adapter pattern for new HTTP source" \
  1.0 true \
  "Pattern was complete - followed steps exactly, tests passed"

# Pattern needed fixes
npx agentdb reflexion store "dp-004" \
  "Used add-stream pattern but needed adjustment" \
  0.6 true \
  "Pattern missing retention field - should update via save-pattern"

Without feedback, the system can't learn which patterns work.

If No Patterns Found

1. Check pattern stats:

   npx agentdb db stats
   

2. Search reflexion episodes:

   npx agentdb reflexion retrieve "your query" --k 10 --synthesize-context
   

3. Check file-based documentation:

   • docs/architecture/ - Architecture documents
   • docs/procedures/ - Step-by-step procedures
   • product/features/*/architecture/ - Feature ADRs

4. After implementing, store the new pattern via save-pattern

The Pattern Workflow

1. BEFORE work:  get-pattern  → Search for relevant patterns (THIS SKILL)
2. DURING work:  Apply the pattern, note what works/gaps
3. AFTER work:   reflexion    → Record if pattern helped (required)
                 save-pattern → Store NEW discoveries (if any)
                 learner      → Auto-discover patterns from episodes (periodic)

Related Skills

• save-pattern - Store NEW patterns after discovering reusable approaches
• reflexion - Record feedback on pattern effectiveness (REQUIRED after using patterns)
• learner - Auto-discover patterns from successful episodes (user-invoked)

What NOT to Use This For

Patterns are PERMANENT application knowledge, not transient swarm state.