AI-augmented work workflow based on Addy Osmani's proven methodology for maximizing LLM effectiveness through structured processes, clear context, and human oversight.
LLMWorkflow implements 10 core principles that treat AI as a powerful pair programmer (not an autonomous agent). The human remains the director; the AI is a capable collaborator.
Adaptable for your use cases:
- Teaching materials (lesson plans, assessments, unit design)
- Content creation (blog posts, documentation, presentations)
- Software development (features, debugging, refactoring)
- Plan Before Coding - Rapid structured planning prevents wasted cycles
- Break Work Into Small Chunks - Small, focused tasks with quick course correction
- Provide Extensive Context - Rich context = better AI output
- Choose the Right Model Strategically - Claude for reasoning, GPT-4 for integration, Gemini for speed
- Leverage AI Agents Across Lifecycle - Use agents for routine tasks with supervision
- Maintain Human Oversight - Review all AI output as if from junior collaborator
- Use Frequent, Granular Commits - Save points enable fearless experimentation
- Customize AI Behavior - Rules files guide AI to your patterns
- Embrace Strong Testing - Automated feedback refines outputs
- Continuous Learning - AI amplifies existing expertise
LLMWorkflow/
├── SKILL.md # Skill routing and overview
├── README.md # This file
├── workflows/
│ ├── Plan.md # Planning workflow (waterfall in 15 min)
│ ├── Implement.md # Implementation workflow (chunked execution)
│ ├── Review.md # Review workflow (quality assurance)
│ ├── Test.md # Testing workflow (verification)
│ ├── Commit.md # Commit workflow (save points)
│ ├── Debug.md # Debug workflow (systematic issue resolution)
│ └── Refactor.md # Refactor workflow (improve without changing behavior)
└── tools/
└── llm-workflow-cli/ # CLI tool for workflow orchestration (Bun/TypeScript)
| Workflow | Triggers | Purpose |
|---|---|---|
| Plan | "plan this work", "create plan for" | Create detailed specs and task breakdowns |
| Implement | "implement this", "build this feature" | Execute work in small, verified chunks |
| Review | "review this work", "check quality" | Systematic quality checking |
| Test | "test this", "verify this works" | Verify correctness and coverage |
| Commit | "commit this work", "save checkpoint" | Create save points for completed work |
| Debug | "debug this", "fix this issue" | Systematically identify and fix issues |
| Refactor | "refactor this", "modernize this" | Improve structure without changing behavior |
User: "Plan a lesson on Knowledge and Language"
-> Plan workflow creates structured lesson plan with chunks
User: "Implement the first chunk"
-> Implement workflow creates opening hook activity
User: "Review this for clarity"
-> Review workflow checks pedagogical soundness
User: "Commit this"
-> Commit workflow saves checkpoint
User: "Plan authentication with JWT"
-> Plan workflow creates detailed spec and task breakdown
User: "Implement chunk by chunk"
-> Implement workflow builds in small, testable pieces
User: "Test this"
-> Test workflow verifies functionality
User: "Commit each chunk"
-> Commit workflow creates save points
User: "Debug memory leak in event handlers"
-> Debug workflow systematically identifies root cause
-> Implement workflow creates fix
-> Test workflow verifies solution
-> Commit workflow saves fix with regression test
CLI-First Pattern:
Goal: Structured AI-augmented work
|
Code: llm-workflow-cli (deterministic workflow orchestration)
|
CLI: plan, implement, review, test, commit, debug, refactor commands
|
Prompts: Workflow files (AI intelligence for content generation)
Integration:
- State Tracking: Update project status in your state directory
- TodoWrite: Track chunk progress visually
- MEMORY System: Save learnings and track completed work
Current Version: 1.0 Status: Production Ready - Workflow files complete Created: 2026-01-16
Completed:
- SKILL.md with proper routing
- 7 comprehensive workflow files
- CLI tool skeleton (Bun/TypeScript)
- System integration
Future Enhancements:
- Full CLI tool implementation
- Configuration system (.llm-workflow.json)
- Context management utilities
- Model switching automation
- Integration with existing tools
Kept exactly as-is:
- 10 core principles
- Workflow phases (Plan -> Implement -> Review -> Test -> Commit)
- Philosophy (AI as collaborator, human as director)
- Best practices and anti-patterns
Customization points:
- Workflow-based routing (adaptable to slash commands)
- Integration with your state/memory system
- TodoWrite progress tracking
- Expandable for teaching content, content creation, or software development
Example adaptations:
- Teaching workflows: add lesson planning, curriculum design, assessment creation
- Content creation: add blog post, documentation, presentation workflows
- Software development: extend with deployment, security review workflows
Original Methodology: Addy Osmani (Google Chrome) Implementation: Based on Addy Osmani's methodology, created 2026-01-16
- Workflow emphasizes human oversight - you're always accountable
- Small chunks enable quick course correction
- Frequent commits create safety net for experimentation
- Review is non-negotiable for quality
- Learning amplifies expertise, doesn't replace it