v4.0+ — Open Source

Transform AI into a
Development Team

AWF is a prompt engineering framework that turns your AI coding assistant into specialized personas — PM, Developer, Designer, Detective — each with structured workflows and safety guardrails.

Get Started View on GitHub
22
Workflows
8
Built-in Skills
7
Personas
5
Shared Protocols
Developer Cycle
From Idea to Production
Every stage of development has a dedicated workflow with its own expert persona.
# New Project /init → /brainstorm → /plan → /design → /code → /code-review → /test → /deploy # Daily Work /recap → /code → /run → /code-review → /test → /save-brain # Bug Fixing /debug → /code-review → /test → /rollback (if needed) # Pre-Release /audit → /code-review → /test → /deploy
Key Features
What Makes AWF Different
Not just prompts — a complete engineering system with process enforcement.
🎭

Persona System

Each workflow has a named expert: PM Ha plans features, Detective Long debugs methodically, Security Eng Khang audits thoroughly. Consistent expertise across sessions.

🛡

Anti-Rationalization

Tables of common AI excuses that are explicitly blocked. "Just a small change, skip tests" → BLOCKED. Applied to 7 workflows.

👥

Non-Tech Mode

Every workflow supports 3 levels: Newbie (plain language, 1 question per turn), Basic, and Technical. Error messages auto-translated to human language.

🧠

Context Management

Never lose context between sessions. Lazy Checkpoint (~20 tokens/task), Proactive Handover when context is 80% full, MCP Memory integration.

🔍

Search-First Coding

Before writing ANY new code, mandatory search for existing solutions. Research → Reuse → Build. Prevents reinventing the wheel.

⚙

Shared Protocols

5 cross-cutting protocols: context-retrieval, iterative-retrieval, resilience (auto-retry), token-discipline, GitHub CLI mapping.

Complete Coverage
22 Workflows
Every stage of software development covered with structured processes.

Development Lifecycle

CommandPersonaRole
/initInitializerCapture idea, create workspace
/brainstormResearch AnalystExplore & validate ideas, create BRIEF.md
/planPM (10yr exp)Features, phases, risk assessment
/designArchitect (15yr)DB schema, API design, acceptance criteria
/visualizeUX DesignerUI mockup generation
/codeSenior Dev (12yr)Search-First implementation + Phase Gates
/testQA EngineerTest strategy, flaky detection, coverage
/debugDetective (8yr)Root cause with Iron Law enforcement
/code-reviewReviewerPre-commit gate, security checklist
/refactorCode GardenerReadability cleanup, safety tiers
/auditSecurity Eng (10yr)Full checkup, A-F grading system

Operations & Utilities

CommandRole
/runAuto-detect build system & launch app
/deployFull production: SEO, analytics, legal, backup, monitoring
/rollbackEmergency recovery via git
/recapContext recovery from .brain/ files
/save-brainPersist knowledge for next session
/helpContext-aware help system
/nextAI suggests next action
/customizePreferences (tech level, tone, autonomy)
/design-systemDesign DNA + UI component library
Process Enforcement
Anti-Rationalization
Inspired by Superpowers. Common AI excuses are explicitly blocked across 7 workflows.
❌
"I already know how to build this"
Search anyway. Existing code, libraries, or patterns save hours of work.
❌
"Just a small change, no test needed"
Small changes cause the worst bugs. Run the test.
❌
"Fix this bug while refactoring"
NEVER mix refactoring and bug fixing. Separate commits, separate concerns.
❌
"Simple project, skip Risk Assessment"
Simple projects fail for simple reasons. Risk Assessment takes 2 minutes.
❌
"Flaky test, just re-run"
Flaky tests hide real bugs. Quarantine and investigate.
❌
"The fix is obvious, skip root cause"
Iron Law: No fix without confirmed root cause. Obvious fixes create new bugs.
Under the Hood
3-Tier Architecture
Agent, Workflow, and Skill layers work together seamlessly.
Tier 1: Agent Layer ← GEMINI.md (global rules)
Tier 2: Workflow Layer ← /plan, /code, /debug...
• Shared Protocols ← resilience, tokens
Tier 3: Skill Layer ← awf-* (auto-activated)
• Custom Skills ← Your own skills
Get Started
Installation
One command. Auto-creates symlinks for .agent/ and .agents/ directories for multi-agent IDE compatibility.
Global Installrecommended
Per-Project
Manual

Install AWF globally so every project benefits. Installs to ~/.gemini/antigravity/ and auto-creates symlinks in ~/.agent/ and ~/.agents/.

curl -fsSL https://raw.githubusercontent.com/devtdq1701/awf/main/install.sh | bash

Creates: ~/.gemini/antigravity/{global_workflows,skills,schemas,templates}
Symlinks: ~/.agent/skills/awf, ~/.agents/skills/awf → primary

Install into current project only. Useful for team repos where workflows are versioned alongside code. Installs to .gemini/ and symlinks .agent/, .agents/.

curl -fsSL https://raw.githubusercontent.com/devtdq1701/awf/main/install.sh | bash -s -- --project

Creates: .gemini/{workflows,skills,schemas,templates}
Symlinks: .agent/skills/awf, .agents/skills/awf → primary

Manual clone & copy if you prefer not to pipe to bash, or want to inspect the files first.

# Clone git clone --depth 1 https://github.com/devtdq1701/awf.git /tmp/awf # Copy to Gemini (global) mkdir -p ~/.gemini/antigravity cp -r /tmp/awf/workflows ~/.gemini/antigravity/global_workflows cp -r /tmp/awf/skills ~/.gemini/antigravity/skills cp -r /tmp/awf/schemas ~/.gemini/antigravity/schemas cp -r /tmp/awf/templates ~/.gemini/antigravity/templates # Symlink for other agent IDEs mkdir -p ~/.agent/skills ~/.agents/skills ln -sf ~/.gemini/antigravity/skills ~/.agent/skills/awf ln -sf ~/.gemini/antigravity/skills ~/.agents/skills/awf # Clean up rm -rf /tmp/awf