save-pattern

name	save-pattern
description	Store APPLICATION patterns (architecture, procedures, conventions) in AgentDB's patterns table. NOT for swarm/transient memory.

Save Pattern - Store Application Knowledge

What This Skill Does

Stores application patterns to AgentDB's patterns table with semantic embeddings. Patterns are searchable via get-pattern using agentdb_pattern_search.

Use this AFTER completing work to share reusable knowledge with future agents.

---

Quick Reference

# Store a new pattern
mcp__agentdb__agentdb_pattern_store(
  taskType="architecture:domain-adapter",
  approach="Description of the pattern and how to apply it...",
  successRate=0.9,
  tags=["architecture", "rust", "ndp"]
)

# Check existing patterns first (avoid duplicates)
mcp__agentdb__agentdb_pattern_search(task="pattern topic", k=3)

# Get pattern statistics
mcp__agentdb__agentdb_pattern_stats()

---

Primary Method: Pattern Store

mcp__agentdb__agentdb_pattern_store(
  taskType="<category:name>",
  approach="<full pattern description>",
  successRate=<0-1>,
  tags=["tag1", "tag2"]
)

Parameters

Parameter	Type	Required	Description
taskType	string	Yes	Category and name (e.g., architecture:domain-adapter)
approach	string	Yes	Full pattern content - what it does, how to use it
successRate	number	No	Confidence level 0-1 (default: 0.9 for proven patterns)
tags	array	No	Array of tags for filtering

---

Examples

Store Architecture Pattern

mcp__agentdb__agentdb_pattern_store(
  taskType="architecture:domain-adapter-source",
  approach="Domain Adapter Pattern for Data Sources: All data sources implement the Source trait for uniform handling. Steps: 1) Create struct implementing Source trait, 2) Implement fetch() -> Vec<TimeSeriesPoint>, 3) Implement health_check() -> HealthStatus. Related files: core/src/traits.rs, core/src/sources/http_poll.rs",
  successRate=0.95,
  tags=["architecture", "hexagonal", "traits", "source"]
)

Store Development Procedure

mcp__agentdb__agentdb_pattern_store(
  taskType="procedure:add-data-stream",
  approach="Add New Data Stream: Prerequisites - Stream config YAML ready, etcd running. Steps: 1) Create config/base/streams/{stream-id}/config.yaml, 2) Define fields array with name, source_path, unit, 3) Run ./deploy.sh sync, 4) Verify: etcdctl get /streams/{id}/config",
  successRate=0.9,
  tags=["procedure", "streams", "config", "etcd"]
)

Store Troubleshooting Pattern

mcp__agentdb__agentdb_pattern_store(
  taskType="troubleshoot:mqtt-data-not-appearing",
  approach="MQTT Data Not Appearing - Symptoms: Sensor data not in Parquet files, no errors in logs. Root Causes: 1) Topic mismatch, 2) Missing stream_id in routing. Solution: 1) Check mosquitto_sub -t # for actual topics, 2) Verify config.yaml source.topics matches, 3) Ensure IngestionRouter tags stream_id",
  successRate=0.85,
  tags=["troubleshoot", "mqtt", "debugging", "parquet"]
)

Store Naming Convention

mcp__agentdb__agentdb_pattern_store(
  taskType="conventions:naming",
  approach="Stream IDs use kebab-case (outdoor-weather, air-quality-pm25). Database fields use snake_case (temperature_celsius). Rust structs use PascalCase (WeatherReading). Config keys use dot notation (streams.outdoor-weather.enabled).",
  successRate=0.95,
  tags=["conventions", "naming", "style"]
)

---

Pattern Categories (taskType)

Use consistent taskType prefixes for categorization:

Category	taskType Prefix	Examples
Architecture	architecture:	architecture:domain-adapter, architecture:data-layers
Procedures	procedure:	procedure:add-stream, procedure:deploy-to-pi
Implementation	implementation:	implementation:etl-persistence, implementation:mcp-tool
Configuration	configuration:	configuration:gitops, configuration:etcd
Testing	testing:	testing:mcp-tool, testing:csv-dimension
Deployment	deployment:	deployment:docker, deployment:resource-constraints
Troubleshooting	troubleshoot:	troubleshoot:mqtt-issues, troubleshoot:parquet-errors
Conventions	conventions:	conventions:naming, conventions:code-style
MCP Tools	mcp:	mcp:tool-implementation, mcp:silver-storage
ETL	etl:	etl:run-lifecycle, etl:persistence
Data Quality	data-quality:	data-quality:framework, data-quality:csv-patterns

---

Best Practices

1. Check First

Always search before creating to avoid duplicates:

mcp__agentdb__agentdb_pattern_search(task="pattern topic", k=5)

2. Be Specific

Include concrete details:

• Good: "Create config/base/streams/{id}/config.yaml with fields array containing name, source_path, unit"
• Bad: "Create a config file"

3. Include Tags

Add relevant tags for better searchability.

4. Reference Files

Mention actual code paths:

"Related files: core/src/traits.rs, docs/procedures/HOW_TO_ADD_STREAM.md"

5. Include Verification

How to confirm the pattern worked:

"Verify: Run cargo test, check logs for 'Source initialized'"

---

Update vs. Create New

Use /pattern-manage for updates and cleanup:

1. Update in place (preferred — no duplicate):

   node .claude/skills/pattern-manage/scripts/pattern-manage.js update 5 --approach "New content..." --success_rate 0.95
   

2. Deprecate old + create new (when approach fundamentally changed):

   node .claude/skills/pattern-manage/scripts/pattern-manage.js deprecate 5
   

   Then use agentdb_pattern_store to create the replacement.

3. Delete obsolete (pattern references deleted code or is a duplicate):

   node .claude/skills/pattern-manage/scripts/pattern-manage.js delete 5
   

4. Find duplicates:

   node .claude/skills/pattern-manage/scripts/pattern-manage.js duplicates
   

---

The Pattern Workflow

1. BEFORE work:  get-pattern  → Search for existing patterns
2. DURING work:  Note gaps, discover new approaches
3. AFTER work:   save-pattern → Store NEW discoveries (THIS SKILL)
                 reflexion    → Record if existing patterns helped
                 learner      → Auto-discover patterns from episodes

---

Related Skills

• get-pattern - Search patterns BEFORE work (always check first)
• reflexion - Record feedback on pattern effectiveness
• learner - Auto-discover patterns from successful episodes
• pattern-manage - Delete, deprecate, update, deduplicate patterns

---

What NOT to Use This For

Don't Store	Use Instead
Swarm coordination state	claude-flow memory tools
Agent task status	claude-flow task tools
Temporary working memory	claude-flow memory with TTL
Session-specific context	claude-flow memory tools
Feedback on patterns	reflexion skill

Patterns are PERMANENT application knowledge, not transient swarm state.