learner

name	learner
description	Auto-discover patterns from reflexion episodes. Run post-feature to consolidate successful approaches into reusable patterns.

Learner - Auto-Discover Patterns

What This Skill Does

Analyzes reflexion episodes to automatically discover:

1. Causal patterns - What actions lead to successful outcomes
2. Reusable patterns - Stored to the patterns table via agentdb_pattern_store
3. Patterns needing review - Low-performing or conflicting patterns

Run this AFTER completing a feature to consolidate learnings.

Note: For manual pattern storage, use save-pattern skill. This skill uses MCP tools exclusively. All discovered knowledge goes to the patterns table (searchable via get-pattern).

---

Quick Reference

# Discover causal patterns from episodes
mcp__agentdb__learner_discover(min_attempts=3, min_success_rate=0.6, min_confidence=0.7)

# Query causal edges
mcp__agentdb__causal_query(min_confidence=0.5, limit=10)

# View database statistics
mcp__agentdb__agentdb_stats(detailed=true)

# Search discovered patterns
mcp__agentdb__agentdb_pattern_search(task="feature-topic", k=5)

---

Step 1: Discover Causal Patterns

Auto-discover causal patterns from reflexion episodes:

mcp__agentdb__learner_discover(
  min_attempts=3,
  min_success_rate=0.6,
  min_confidence=0.7,
  dry_run=false
)

Parameters

Parameter	Type	Default	Description
min_attempts	number	3	Minimum times pattern was tried
min_success_rate	number	0.6	Minimum success rate
min_confidence	number	0.7	Statistical confidence threshold
dry_run	boolean	false	Preview without storing

Examples

Standard discovery:

mcp__agentdb__learner_discover(min_attempts=3, min_success_rate=0.6, min_confidence=0.7)

Aggressive (more patterns, lower thresholds):

mcp__agentdb__learner_discover(min_attempts=2, min_success_rate=0.5, min_confidence=0.6)

Conservative (fewer, higher-confidence patterns):

mcp__agentdb__learner_discover(min_attempts=5, min_success_rate=0.8, min_confidence=0.9)

Dry run (preview without storing):

mcp__agentdb__learner_discover(min_attempts=3, min_success_rate=0.6, min_confidence=0.7, dry_run=true)

---

Step 2: Store Discovered Patterns

After learner_discover returns results, store high-value discoveries to the patterns table so get-pattern can find them:

mcp__agentdb__agentdb_pattern_store(
  taskType="learner:discovered-pattern-name",
  approach="Description of the discovered pattern, including cause-effect relationship and how to apply it",
  successRate=0.85,
  tags=["learner", "auto-discovered", "topic"]
)

This replaces the legacy agentdb skill consolidate command, which wrote to the skills table (orphaned from get-pattern searches).

---

Step 3: Query What Was Learned

View Causal Edges

mcp__agentdb__causal_query(min_confidence=0.5, limit=10)

With filters:

# Filter by cause
mcp__agentdb__causal_query(cause="Source trait", min_confidence=0.7, min_uplift=0.1, limit=20)

# Filter by effect
mcp__agentdb__causal_query(effect="data ingestion", min_confidence=0.8, limit=10)

Search Patterns (preferred)

mcp__agentdb__agentdb_pattern_search(task="data ingestion", k=5)

Search Legacy Skills (read-only)

mcp__agentdb__skill_search(task="data ingestion", k=5)

View Database Stats

mcp__agentdb__agentdb_stats(detailed=true)

---

Prune Low-Quality Data (CLI Only)

These operations have no MCP equivalent. Use CLI when needed:

# Remove episodes older than 90 days with reward < 0.5
agentdb reflexion prune 90 0.5

# Remove low-confidence causal edges
agentdb learner prune 0.5 0.05 90

---

Post-Feature Workflow

Run after completing a feature:

# 1. Discover causal patterns
mcp__agentdb__learner_discover(min_attempts=3, min_success_rate=0.7, min_confidence=0.8)

# 2. Review results, then store valuable patterns to patterns table
mcp__agentdb__agentdb_pattern_store(
  taskType="learner:pattern-name",
  approach="What was discovered and how to apply it",
  successRate=0.85,
  tags=["learner", "auto-discovered"]
)

# 3. View what was learned
mcp__agentdb__agentdb_stats(detailed=true)

# 4. Search patterns to verify they're findable
mcp__agentdb__agentdb_pattern_search(task="feature-topic", k=5)

---

Understanding Results

Causal Edges (from learner_discover)

Cause: "Using Source trait with health_check"
Effect: "Reliable data ingestion with automatic recovery"
Uplift: 0.35 (35% improvement)
Confidence: 0.92

Stored Patterns (from agentdb_pattern_store)

taskType: "learner:http-source-reliability"
approach: "Implementing health_check on Source trait leads to 35% more reliable ingestion..."
successRate: 0.89
tags: ["learner", "auto-discovered", "source-trait"]

---

Thresholds Guide

Parameter	Low (exploratory)	Standard	High (conservative)
min_attempts	2	3	5
min_success_rate	0.5	0.7	0.9
min_confidence	0.6	0.8	0.95

---

Maintenance Schedule

Frequency	Action	Tool
Post-feature	Discover patterns	mcp__agentdb__learner_discover
Post-feature	Store to patterns table	mcp__agentdb__agentdb_pattern_store
Monthly	Review stats	mcp__agentdb__agentdb_stats
Quarterly	Prune stale data	agentdb reflexion prune (CLI)

---

The Pattern Workflow

1. BEFORE work:  get-pattern  → Search for relevant patterns
2. DURING work:  Apply patterns, note gaps
3. AFTER work:   reflexion    → Record what helped
                 save-pattern → Store NEW discoveries manually
                 learner      → Auto-discover patterns (THIS SKILL)

---

Related Skills

• get-pattern - Search patterns BEFORE work
• save-pattern - Store NEW patterns manually
• reflexion - Record feedback that feeds learner

---

What NOT to Use This For

Don't Use For	Use Instead
Storing specific patterns	save-pattern
Recording work feedback	reflexion
Searching patterns	get-pattern

Learner is for AUTOMATIC discovery, not manual pattern management.