Transform your AI coding assistant into a structured software development team.
AWF is a prompt engineering framework that turns AI assistants (Gemini, Claude, etc.) into specialized personas — PM, Developer, Designer, Detective, Auditor — each with dedicated workflows, checklists, and safety guardrails.
Problem: AI coding assistants answer correctly but lack process — no code review, no testing, no edge case handling, no context persistence.
Solution: AWF wraps every task in a structured workflow with phases, checklists, and anti-rationalization guards.
AI PROPOSES → User APPROVES → AI EXECUTES → AI SELF-CHECKS
| Principle | Description |
|---|---|
| AI-First Proposal | AI proposes solutions first, user only approves/adjusts |
| Persona-Driven | Each workflow has a dedicated expert persona |
| Context Never Lost | Lazy Checkpoint + Proactive Handover = no lost context between sessions |
| Non-Tech Friendly | Every workflow supports non-technical users with plain language |
| Anti-Rationalization | Tables of common AI excuses that are explicitly blocked |
~/.gemini/
├── GEMINI.md # Global rules (your customization)
├── antigravity/
│ ├── global_workflows/ # 22 workflow files
│ │ └── _shared/ # 5 shared protocols
│ ├── skills/ # 8 built-in skills
│ ├── schemas/ # JSON schemas for .brain files
│ └── templates/ # Example configs
┌─────────────────────────────┐
│ Tier 1: Agent Layer │ ← GEMINI.md (global rules, persona)
├─────────────────────────────┤
│ Tier 2: Workflow Layer │ ← /brainstorm, /plan, /code, /debug...
│ └── Shared Protocols │ ← _shared/ (resilience, token-discipline)
├─────────────────────────────┤
│ Tier 3: Skill Layer │ ← awf-* skills (auto-activated)
│ └── Custom Skills │ ← Your own skills
└─────────────────────────────┘
| Command | Persona | Role |
|---|---|---|
/init |
Project Initializer | Capture idea, create workspace |
/brainstorm |
Research Analyst | Explore & validate ideas, create BRIEF.md |
/plan |
PM (10yr exp) | Features, phases, risk assessment |
/design |
Architect (15yr exp) | DB schema, API design, acceptance criteria |
/visualize |
UX Designer | UI mockup generation |
/code |
Senior Dev (12yr exp) | Implementation with Search-First + Phase Gates |
/test |
QA Engineer | Test strategy, flaky detection |
/debug |
Detective (8yr exp) | Root cause investigation with Iron Law |
/code-review |
Code Reviewer | Pre-commit gate, security checklist |
/refactor |
Code Gardener | Readability cleanup, safety tiers |
/audit |
Security Eng (10yr exp) | Full checkup, A-F grading |
| Command | Role |
|---|---|
/run |
Auto-detect & launch app |
/deploy |
Full production guide (SEO, analytics, legal, backup, monitoring) |
/rollback |
Emergency recovery |
/recap |
Context recovery from .brain/ |
/save-brain |
Persist knowledge for next session |
/review |
Project overview & handover |
| Command | Role |
|---|---|
/help |
Context-aware help |
/next |
AI suggests next action |
/customize |
Preferences management |
/design-system |
Design DNA + UI component library |
/gen-dockerfile |
Secure Docker Architect |
# New project
/init → /brainstorm → /plan → /design → /code → /test → /deploy
# Daily work
/recap → /code → /run → /test → /save-brain
# Bug fixing
/debug → /test → /rollback (if needed)
# Pre-release
/audit → /code-review → /test → /deploy → /save-brainEach workflow has a named persona with specific traits:
| Persona | Workflow | Style |
|---|---|---|
| PM | /plan | "Do less, do well", offers 2-3 options |
| Architect | /design | Explains tech in plain language |
| Senior Dev | /code | Careful, explains reasoning |
| Detective | /debug | Calm, methodical, never rushes |
| Security Eng | /audit | Meticulous, pairs problems with solutions |
Read from preferences.json, supports 3 levels:
- Newbie: Hide technical details, emoji errors, 1 question per turn
- Basic: Text + diagrams with explanations
- Technical: Full technical output
Every workflow includes an error translation table for non-tech users.
Anti-Rationalization (inspired by Superpowers)
Tables of common AI excuses that are explicitly blocked:
| Excuse | Reality |
| ------ | ------- |
| "Simple issue, skip process" | Simple bugs have root causes too. |
| "I already know how to build this" | Search anyway. Existing code saves hours. |
| "Just a small change, no test needed" | Small changes cause the worst bugs. |Applied to 7 workflows: /brainstorm, /plan, /code, /test, /debug, /code-review, /refactor.
.brain/
├── brain.json # Static project knowledge
├── session.json # Dynamic session state
└── preferences.json # User preferences
- Lazy Checkpoint: ~20 tokens per task, ~450 per phase
- Proactive Handover: Auto-save when context >80% full
- Cross-session Recovery: MCP Memory → Knowledge Items → Conversation Logs
| Protocol | Purpose |
|---|---|
context-retrieval.md |
Memory → KI → Logs priority order |
iterative-retrieval.md |
Large codebase search (30+ files) |
resilience.md |
Auto-retry, timeout, fallback patterns |
token-discipline.md |
Context window optimization |
gh-cli.md |
GitHub CLI ↔ MCP tool mapping |
| Skill | Purpose |
|---|---|
awf-adaptive-language |
Adjust terminology by user level |
awf-auto-save |
Auto-save session context |
awf-context-help |
Context-aware help |
awf-error-translator |
Translate errors to plain language |
awf-onboarding |
First-time user guide |
awf-session-restore |
Lazy-loading context restore |
context-minify |
Minify source for ~40-60% token savings |
- Copy the
workflows/directory to~/.gemini/antigravity/global_workflows/ - Copy the
skills/directory to~/.gemini/antigravity/skills/ - Copy
schemas/andtemplates/to their respective directories - Start using: type
/helpin your AI assistant
# Clone the repo
git clone https://github.com/devtdq1701/awf.git /tmp/awf-install
# Copy to Antigravity directory
mkdir -p ~/.gemini/antigravity
cp -r /tmp/awf-install/workflows ~/.gemini/antigravity/global_workflows
cp -r /tmp/awf-install/skills ~/.gemini/antigravity/skills
cp -r /tmp/awf-install/schemas ~/.gemini/antigravity/schemas
cp -r /tmp/awf-install/templates ~/.gemini/antigravity/templates
# Clean up
rm -rf /tmp/awf-install
echo "AWF installed! Type /help to get started."# Initialize preferences
/customize
# Create your first project
/initCreate a new .md file in workflows/:
---
description: My Custom Workflow
---
# WORKFLOW: /my-workflow - The [Role Name]
You are **[Persona Name]**. [Mission description].
## Phase 1: [Phase Name]
...
## Anti-Rationalization
| Excuse | Reality |
...
## Next Steps
...Create a directory in skills/ with a SKILL.md:
skills/my-skill/
├── SKILL.md # Main instructions (required)
├── scripts/ # Helper scripts
└── examples/ # Reference implementations
- AWF Original by @TUAN130294 — Original creator of the Antigravity Workflow Framework. This repo is a fork with enhancements (Anti-Rationalization, Superpowers-inspired improvements, English translation, additional docs).
- Superpowers — Anti-Rationalization patterns, Iron Law debugging, structured feedback protocols
- BMAD Method — Persona-driven workflow design
MIT License — Feel free to use, modify, and share.
AWF — Your dreams, our engineering.