Turn raw AI capability into reliable engineering output.
FORGE is a structured methodology for integrating AI into real engineering workflows. It provides universal thinking patterns, pluggable domain standards, AI personas, and self-improving feedback loops — all designed to make AI-assisted development repeatable, governed, and production-grade.
Created by Chase Logan — built from real-world experience deploying AI across industrial automation, legacy modernization, and enterprise software engineering.
Most teams use AI as a chatbot. FORGE treats AI as an engineering tool with:
- Structured workflows — not ad-hoc prompting
- Domain-specific standards — not generic suggestions
- Governance and verification — not blind trust
- Measurable outcomes — not anecdotal wins
Real-world results:
- 21x developer efficiency improvement (measured across production deployments)
- $0.002/line cost for trained AI workflows vs $0.014/line untrained
- 56x documentation speed improvement
- 240x standards-checking acceleration
| Problem Type | Workflow | When to Use |
|---|---|---|
| Build something new | R-I-S-E | Creation, design, new features |
| Fix or optimize | C-A-R-E | Bugs, refactoring, performance |
| Understand or document | HARVEST | Documentation, research, analysis |
Copy the content of FORGE_MASTER.md into any AI assistant (Claude, ChatGPT, Gemini, Cursor, Windsurf) to activate the full framework.
"Create a REST API for inventory management with proper error handling"
- Classifies your problem automatically
- Selects the optimal workflow
- Applies domain standards
- Verifies the solution before delivery
FORGE/
├── README.md # You are here
├── FORGE_MASTER.md # Complete framework — paste into any AI
├── QUICK_START.md # 5-minute getting started guide
│
├── workflows/ # HOW to think
│ ├── rise.md # Research → Implement → Synthesize → Execute
│ ├── care.md # Collect → Analyze → Refine → Execute
│ ├── harvest.md # Documentation & understanding
│ ├── omega_loop.md # Self-improving feedback loop
│ ├── problem_classifier.md # Automatic problem categorization
│ ├── verification.md # 5-layer correctness stack
│ ├── tree_of_thoughts.md # Multi-path exploration
│ └── react_integration.md # Reasoning + Acting loops
│
├── personas/ # WHO does the work
│ ├── atlas.md # Deep Research specialist
│ ├── sage.md # Architecture & system design
│ ├── scribe.md # Technical documentation
│ └── sentinel.md # Security testing
│
├── core/ # Enhancement layers
│ ├── meta_prompting.md # Self-critique before delivery
│ ├── confidence_protocol.md # Explicit confidence scoring
│ ├── assumption_tracker.md # Track and validate assumptions
│ └── orchestration.md # Multi-agent coordination
│
├── modules/ # WHAT standards to follow
│ ├── domain_module_template.md # Template for creating new modules
│ ├── web_development.yaml # React, TypeScript, FastAPI
│ ├── python_data.yaml # Python, pandas, ML
│ ├── general_reasoning.yaml # Logic, decisions, analysis
│ ├── research_analysis.yaml # Research methodology
│ └── security_testing.yaml # Security assessment standards
│
├── execution/ # Runtime engine
│ ├── session_manager.md # Persistent state management
│ ├── checkpoint_manager.md # Git-based checkpointing
│ ├── audit_system.md # Crash-safe logging
│ ├── error_handling.md # Retry logic & error categories
│ └── parallel_agents.md # Multi-agent orchestration
│
├── templates/ # Reusable document templates
│ ├── rise/ # R-I-S-E phase templates
│ ├── care/ # C-A-R-E phase templates
│ ├── harvest/ # Documentation templates
│ └── security/ # Security report templates
│
├── .cursorrules-templates/ # IDE governance rules
│ └── base.cursorrules # Base Cursor IDE rules
│
└── docs/ # Deep documentation
├── theory/ # Mathematical foundations
└── examples/ # Usage examples
┌────────────┐ ┌────────────┐ ┌────────────┐ ┌────────────┐
│ CLASSIFY │ ──→ │ EXECUTE │ ──→ │ VERIFY │ ──→ │ LEARN │
│ Problem │ │ Workflow │ │ Solution │ │ (OMEGA) │
└────────────┘ └────────────┘ └────────────┘ └────────────┘
│ │ │ │
▼ ▼ ▼ ▼
┌────────────┐ ┌─────────────────┐ ┌────────────┐ ┌────────────┐
│ 5 Categories│ │ R-I-S-E │ │ 5 Layers │ │ Knowledge │
│ + Domain │ │ C-A-R-E │ │ Error → 0 │ │ Base │
│ + Complexity│ │ HARVEST │ └────────────┘ └────────────┘
└────────────┘ └─────────────────┘
For creation and transformation problems. Thorough, research-first approach.
| Phase | Purpose | Output |
|---|---|---|
| Research | Understand requirements, constraints, prior art | Research brief |
| Implement | Build the solution with domain standards | Working code/artifact |
| Synthesize | Integrate, test, refine | Verified deliverable |
| Execute | Deploy, document, hand off | Production-ready output |
For repair and optimization problems. Fast, iterative approach.
| Phase | Purpose | Output |
|---|---|---|
| Collect | Gather context, reproduce issue | Problem statement |
| Analyze | Root cause analysis | Diagnosis |
| Refine | Develop and test fix | Verified solution |
| Execute | Apply fix, verify no regressions | Resolved issue |
For understanding problems. Systematic knowledge extraction.
Extracts structured documentation from any codebase, system, or domain — producing layered output from executive summary to implementation details.
Specialized AI personalities that integrate FORGE workflows:
| Persona | Role | Methodology | Best For |
|---|---|---|---|
| Atlas | Deep Research | HARVEST + ReAct | Comprehensive research with sources |
| Sage | Architecture | R-I-S-E + Tree of Thoughts | System design with trade-off analysis |
| Scribe | Documentation | HARVEST + Meta-Prompting | Technical writing for any audience |
| Sentinel | Security | R-I-S-E + C-A-R-E | Vulnerability identification |
Each persona is a self-contained prompt — paste into any AI assistant to activate.
Pluggable standards that inject domain-specific rules into any workflow:
| Module | Domain | Key Standards |
|---|---|---|
web_development |
React/TypeScript/FastAPI | Type safety, error handling, accessibility |
python_data |
Python/ML/Data Science | Type hints, reproducibility, validation |
general_reasoning |
Decisions & analysis | Assumptions, multiple perspectives |
research_analysis |
Research methodology | Source citation, limitations |
security_testing |
Security assessments | OWASP methodology, CVSS scoring |
Create your own: Use modules/domain_module_template.md to build domain modules for your specific tech stack, industry, or codebase.
Every task feeds back into the knowledge base:
OBSERVE → MODEL → EXECUTE → GENERATE → ANALYZE → LEARN → (repeat)
Key property: Error rate converges to zero over iterations.
Error(n) = Error(0) × (1 - learning_rate)^n
After 5 iterations: 40% error → 13% error
After 10 iterations: 40% error → 4% error
After 20 iterations: 40% error → 0.5% error
Layer 5: Human Judgment → Final arbiter
Layer 4: Ensemble → Multiple methods agree (99%+)
Layer 3: Statistical → Benchmarks, metrics (90%+)
Layer 2: Automated Testing → Tests pass (95%+)
Layer 1: Formal → Types, syntax, linting (100%)
Combined: 99.9%+ correctness for well-defined problems
Generate comprehensive, structured documentation for any codebase using GPT-4 and the HARVEST workflow. Three-panel dark industrial UI with animated pipeline visualization.
FORGE includes .cursorrules templates for governing AI behavior directly in your IDE:
# .cursorrules example
framework: FORGE
workflow: auto-detect
domain_module: web_development
confidence_threshold: 7
verification: enabled
persona: sageWorks with Cursor, Windsurf, and any AI-assisted IDE that supports rules files.
- Read
QUICK_START.md— 5 minutes - Load
FORGE_MASTER.mdinto your AI assistant - Pick a persona from
personas/for specialized work - Create a domain module for your tech stack using the template
Apache 2.0 — Use it, extend it, build on it.
FORGE was developed by Chase Logan through hands-on experience deploying AI-driven engineering workflows across industrial automation and enterprise software. It represents a distilled, domain-agnostic methodology for making AI a reliable engineering partner — not just a code generator.
- GitHub: github.com/0xPliny
- Framework: FORGE v1.0
FORGE: Because AI without governance is just expensive autocomplete.