/moe-learn

Train the MoE classification system from confirmed classifications and user feedback.

Usage

# Confirm a correct classification
/moe-learn --confirm <path> <actual_type>

# Confirm multiple from file
/moe-learn --batch <confirmations.json>

# Show pending confirmations
/moe-learn --pending

# Show learning statistics
/moe-learn --stats

# Export training data
/moe-learn --export <output.json>

# Bootstrap from frontmatter
/moe-learn --bootstrap <docs_path>

Arguments

Argument	Description
--confirm <path> <type>	Confirm classification for a specific document
--batch <file>	Process batch confirmations from JSON file
--pending	Show classifications awaiting confirmation
--stats	Show learning system statistics
--export <file>	Export training data to JSON
--bootstrap <path>	Bootstrap from documents with frontmatter types

System Prompt

You are the MoE Learning Manager. Execute the requested learning operation:

1. For --confirm: Record that a document was correctly or incorrectly classified

   • Update analyst accuracy scores
   • Recalculate dynamic weights
   • Update calibration data

2. For --batch: Process multiple confirmations from a JSON file with format:

   [{"path": "docs/file.md", "actual_type": "guide"}, ...]

3. For --pending: List classifications that haven't been confirmed yet

4. For --stats: Show:

   • Total classifications recorded
   • Confirmation rate
   • Overall accuracy
   • Top/bottom analysts by accuracy

5. For --export: Export all training data for analysis

6. For --bootstrap: Scan directory for documents with frontmatter type declarations and use as ground truth

Execute by running:

import sys
sys.path.insert(0, 'scripts/moe_classifier')
from core.learning import get_learner

learner = get_learner()

# For --confirm:
learner.confirm_classification(document_path, actual_type)

# For --stats:
stats = learner.get_stats()

# For --pending:
pending = learner.get_pending_confirmations()

# For --bootstrap:
count = learner.bootstrap_from_frontmatter(docs_path)

Confirmation Format

When confirming classifications, the actual type must be one of:

• agent - AI agent definitions
• command - Slash commands
• skill - Reusable skills
• guide - How-to guides
• reference - Reference documentation
• workflow - Workflow definitions
• adr - Architecture Decision Records
• config - Configuration files
• hook - Event hooks
• script - Automation scripts

Examples

# Confirm correct classification
/moe-learn --confirm docs/guides/SETUP.md guide

# Correct a misclassification
/moe-learn --confirm agents/reviewer.md agent

# Check what needs confirmation
/moe-learn --pending

# See learning progress
/moe-learn --stats

# Bootstrap from existing typed docs
/moe-learn --bootstrap docs/

Related Commands

• /moe-calibrate - Threshold and calibration management
• /moe-workflow - Full classification pipeline
• /classify - Classify documents

Success Output

When moe-learn completes:

✅ COMMAND COMPLETE: /moe-learn
Action: <confirm|batch|pending|stats|export|bootstrap>
Confirmed: N classifications
Accuracy: N%
Pending: N awaiting

Completion Checklist

Before marking complete:

• Operation identified
• Data processed
• Feedback recorded
• Stats updated

Failure Indicators

This command has FAILED if:

• ❌ No operation specified
• ❌ Invalid type provided
• ❌ Path not found
• ❌ Learning data corrupted

When NOT to Use

Do NOT use when:

• Just classifying (use /classify)
• Managing thresholds (use /moe-calibrate)
• Full pipeline (use /moe-workflow)

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
Wrong type	Bad learning	Verify correct type
Skip pending	No feedback	Review pending regularly
No bootstrap	Slow learning	Bootstrap from frontmatter

Principles

This command embodies:

• #9 Based on Facts - Learn from ground truth
• #6 Clear, Understandable - Clear learning stats
• #3 Complete Execution - Full learning cycle

Full Standard: CODITECT-STANDARD-AUTOMATION.md