google antigravity agent scaffolder

I have architected a two-agent audit-and-fix pipeline. This setup uses a structured ReviewState object to ensure Agent B receives actionable security findings without the noise of the raw audit logs.

Antigravity Multi-Agent Code Review Scaffolding

import json

# 1. State Definition
class ReviewState:
    def __init__(self):
        self.vulnerabilities = [] # List of {severity, loc, description}
        self.code_context = ""
        self.audit_complete = False

# 2. Agent A: Security Auditor
# Persona: Expert AppSec Engineer
# Output Schema: Strict JSON array of vulnerability objects
AUDITOR_SCHEMA = {
    "type": "object",
    "properties": {
        "findings": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "severity": {"type": "string", "enum": ["LOW", "MEDIUM", "HIGH", "CRITICAL"]},
                    "line": {"type": "integer"},
                    "issue": {"type": "string"}
                },
                "required": ["severity", "issue"]
            }
        }
    }
}

# 3. Handoff Logic
def coordinate_review(source_code):
    state = ReviewState()
    state.code_context = source_code
    
    # Trigger Auditor
    raw_audit = execute_agent("Security_Auditor", {"code": source_code, "schema": AUDITOR_SCHEMA})
    state.vulnerabilities = json.loads(raw_audit).get("findings", [])
    state.audit_complete = True
    
    # Structured Handoff to Agent B (Fixer)
    if state.vulnerabilities:
        handoff_payload = {
            "target_code": state.code_context,
            "issues_to_fix": state.vulnerabilities,
            "instruction": "Provide diffs for each finding."
        }
        return execute_agent("Remediation_Expert", handoff_payload)
    
    return "No vulnerabilities found."

Next steps

1. Register the AUDITOR_SCHEMA in your Antigravity Agent Configuration panel.
2. Implement the execute_agent wrapper to interface with your specific Antigravity runtime environment.
3. Add a Logger tool to Agent A to persist raw findings to your database before the handoff.