Antigravity made your agent autonomous. Keep it from nuking prod.

Autonomy is the point — and the risk

An agent that can rm -rf, git push --force, run a migration, or DROP TABLE on its own will eventually do one of those at the wrong moment. Rollback, insurance, and observability all kick in after the damage. Black_Wall runs before the action: your Antigravity agent calls one tool, forecast, and gets back a risk score (0–100), the red flags it found, and a GO / CONFIRM / STOP gate — in time to stop.

Add it to Antigravity in ~60 seconds

1. Get a free key at blackwalltier.com → Dashboard → API keys (~100 forecasts/mo, no card).
2. In Antigravity: agent panel → Manage MCP Servers → View raw config.
3. Add this to mcp_config.json and reload:

{
  "mcpServers": {
    "blackwall": {
      "command": "npx",
      "args": ["-y", "blackwall-mcp"],
      "env": { "BLACKWALL_API_KEY": "bw_live_your_key_here" }
    }
  }
}

Then tell the agent, in your Antigravity rules: “Before any irreversible action, call the forecast tool first and stop if it returns STOP.”

What it catches

• Destructive shell — rm -rf on the wrong path, git push --force over someone's work
• Destructive SQL — DROP/TRUNCATE, or a DELETE/UPDATE with no WHERE
• Prod from a non-prod context — a “clean up staging” task pointed at production
• Prompt injection in tool inputs — poisoned content steering the agent's real action
• Leaking secrets or PII in a commit, message, or external call
• Irreversible with no backup — deletes when no snapshot exists

Each maps to a named red-flag code from the 28-mode taxonomy — so the agent gets a reason, not just a refusal.

Try it with zero risk — observe mode

Not ready to let a guardrail block your agent? Add "BLACKWALL_MODE": "observe" to the config. It scores and logs every action but never blocks — your agent behaves exactly as today. Review your dashboard after a week to see what it would have caught, then switch to enforce.