Knowledge Base Skill (Mega-Skill)

Version: 1.0.0 Purpose: Organizational learning from CFN Loop execution Status: Production Consolidates: workflow-codification, cfn-playbook Confidence: 7.0/10 (dual learning systems)

Overview

This mega-skill provides organizational learning:

• Workflow - Track edge cases, failures, cost metrics, ROI
• Playbook - Store successful patterns, agent configs, iteration strategies

Directory Structure

knowledge-base/
├── SKILL.md              # This file
├── execute.sh            # Main entry point
├── cli/
│   └── knowledge-base.sh # Unified CLI interface
├── lib/
│   ├── workflow/         # From workflow-codification
│   └── playbook/         # From cfn-playbook
└── data/                 # Database files (created on init)
    ├── workflows.db
    ├── playbooks.db
    └── learnings.db

Learning System

• Successes → Playbook (what worked)
• Failures → Workflow codification (what to avoid)
• Combined: Complete organizational memory

Migration Paths

Usage

Main Entry Point

# Initialize databases
./execute.sh init

# Query for patterns
./execute.sh query 'authentication'

# Store new learning
./execute.sh store playbook '{"task_type": "auth", "pattern": "..."}'

# Show help
./execute.sh help

Advanced CLI Usage

# Direct CLI access
./cli/knowledge-base.sh --help

# Query workflow patterns
./cli/knowledge-base.sh query-workflow --pattern 'auth'

# Query playbook entries
./cli/knowledge-base.sh query-playbook --task-type bugfix

# Store learning with metadata
./cli/knowledge-base.sh store-learning \
  --type workflow \
  --category edge-case \
  --data '...' \
  --confidence 0.85

Version History

1.1.0 (2025-12-08)

• Fixed bootstrap utilities path to use shared location
• Created unified CLI interface
• Added main execute.sh entry point
• Integrated workflow and playbook functionality

1.0.0 (2025-12-02)

• Consolidated workflow + playbook into unified knowledge base

Learnings Integration

A lightweight per-project learnings system complements the knowledge base. While the knowledge base stores structured workflow and playbook data in SQLite, the learnings system uses append-only JSONL files for fast, low-friction logging of patterns, pitfalls, and preferences discovered during agent work.

See [LEARNINGS.md](./LEARNINGS.md) for the full specification, including storage format, subcommands (/learn), auto-logging hooks, and pruning rules.