Agent-First Era Protocol

Stop Coding.
Start Architecting.

Google Antigravity shifts the paradigm. Your job is no longer to write loops, but to orchestrate agents safely.

Launch Tools Read Guidelines
~/

Safety

Prevent rm -rf disasters by strictly controlling Terminal Permissions.

⌘

Consistency

Abandon "Fast Mode." Force agents to Plan, Think, and Architect before execution.

◉

Accuracy

Don't trust text logs. Verify with generated Artifacts and Browser Screenshots.

01_TERMINAL_SAFETY

The agent is powerful, but blind. Giving it "Turbo" access is like giving a toddler a flamethrower.

TURBO
Unrestricted root access. High risk of data loss.
AUTO
Pauses for critical commands. The balanced choice.
OFF
Manual approval for everything. Production safe.

02_PLANNING_MODE

Speed is the enemy of structure. Toggling "Planning Mode" forces the agent to generate a Task List Artifact first.

# The Old Way (Fast Mode)
- Agent writes spaghetti code immediately.
- Hallucinates dependencies.

# The Antigravity Way (Planning Mode)
+ Generates Implementation Plan.
+ Reads .cursorrules.
+ Awaits user architectural approval.

03_VERIFICATION_LOOP

Spend your time reviewing plans, not debugging syntax.

  • 1 Read the Plan Artifact. Edit it like a Google Doc.
  • 2 Ask for Browser Screenshots to verify UI.
  • 3 Only approve code execution after visual proof.

04_EXECUTION_PIPELINE

The "Git Flow" for autonomous agents.

Draft Master Prompt

Define the architecture. Don't ask for code yet. Ask for a structure.

> "Build a Python honey inventory system..."

Freeze Scope

Review the Task List Artifact. Delete unnecessary features (scope creep) before approving.

Step-by-Step Exec

Approve one task. Verify output. Then proceed. Never "Run All".

> Task 1: DONE
> Task 2: WAITING...

05_AI_CONSOLE

Powered by Google Gemini 2.5 Flash

This repository does not include a preconfigured API key — you must obtain one and paste it above.

Get a Gemini API key (signup required)
risk_analyzer.exe

# Enter a terminal command to evaluate safety risk

$
agent_planner.py

# Describe project to generate Task Artifact

>>>