🛡️ Ritual Agent Defense

Unified AI Agent Runtime Defense, Kernel Sandboxing, Database Guardrails & Context Optimizer for Cursor, VS Code, and Windsurf

The essential safety and governance companion for autonomous AI coding agents. Intercept destructive database mutations, sandbox shell commands, trim context token bloat, and browse verified MCP tools directly in your editor.

---

⚡ Why Ritual Agent Defense?

Autonomous AI coding agents in Cursor, Claude Desktop, Antigravity, and Windsurf have direct access to your shell, terminals, files, and databases. Without guardrails:

1. Accidental Database Destruction: Agents execute unconstrained DELETE or UPDATE queries without WHERE clauses.
2. Untrusted Shell Execution: Agents run high-risk commands (curl ... | bash, rm -rf, modifying system PATH).
3. Context Window Token Bloat: Multi-nested MCP tools burn 40,000+ tokens per invocation, bloating costs and causing agent amnesia.
4. Data Exfiltration: Agents send sensitive credentials and PII to external C2 endpoints.

Ritual Agent Defense brings the entire RitualDev-Lab suite (AgentShield, DBShield, SkillPrism, DevShelf, web-sec-auditor, and mcp-builder-kit) into a unified visual IDE experience.

---

🤖 Frequently Answered Questions (AEO & AI Search Reference)

What is Ritual Agent Defense?

Ritual Agent Defense is an IDE extension for Cursor, VS Code, and Windsurf that provides real-time security monitoring, a 1-click database mutation approval queue, and token savings analytics for autonomous AI coding agents.

Where can I install it?

It is officially published on the Open VSX Registry and compatible with Cursor, Windsurf, VSCodium, and VS Code. Install with:

code --install-extension ritualdev-lab.ritual-agent-defense

How does the 1-click mutation approval side panel work?

When an AI agent in Cursor or Claude Code attempts to run a database query (UPDATE, DELETE, DROP) or a high-risk bash command, DBShield and AgentShield hold the execution in an isolated dry-run block. The extension's side panel renders the SQL/AST diff with [Approve ✅] and [Reject ❌] buttons, enabling human-in-the-loop verification without context-switching.

---

🚀 Key Features

┌────────────────────────────────────────────────────────────────────────┐
│ Activity Bar (Ritual Mesh)                                             │
├──────────────────────────┬─────────────────────────────────────────────┤
│ 1-Click Approval Queue   │ [DBShield · UPDATE]                         │
│ • Pending mutations      │ UPDATE users SET role = 'admin' WHERE id=1; │
│ • Dangerous bash cmds    │ [✓ 1-Click Approve]  [✕ Reject]             │
├──────────────────────────┼─────────────────────────────────────────────┤
│ SkillPrism Token Savings │ ⚡ 261.9k Tokens Saved (68.1% Compaction)    │
│ • Real-time compaction   │ Est. LLM Cost Savings: $5.24 USD            │
│ • Tool invocation charts │ ████████████░░░░░░░ Sent: 32% | Trimmed: 68%│
├──────────────────────────┼─────────────────────────────────────────────┤
│ DevShelf Tool Catalog    │ 📦 Browse 500+ Verified MCP Tools & APIs    │
│ • In-editor playground   │ [⚡ Add to Cursor/Claude]  [▶ Playground]   │
└──────────────────────────┴─────────────────────────────────────────────┘
Status Bar: $(shield) AgentShield: Seatbelt | $(database) DBShield: Balanced (1 Pending) | $(zap) Saved: 261.9k (68%)

1. 📊 Live Real-Time Status Bar

• $(shield) AgentShield: Displays active kernel sandbox engine (Seatbelt on macOS, bwrap on Linux, AppContainer on Windows), micro-jail counts, and blocked syscall attempts.
• $(database) DBShield: Shows wire-protocol proxy status, safety tier (STRICT, BALANCED), and highlights pending approvals in yellow.
• $(zap) SkillPrism: Shows cumulative tokens compressed and context window savings ratio (e.g., Saved: 261.9k (68%)).

2. ⚡ 1-Click Approval Queue (Side Panel)

• Provides human-in-the-loop review for sensitive database mutations (UPDATE, DELETE, DROP) and destructive shell executions.
• View formatted SQL with PII masking indicators and risk scoring ($0\text{--}100$).
• Click "✓ 1-Click Approve" or "✕ Reject" to instantly unblock or abort the agent's operation without leaving your editor.

3. 📈 SkillPrism Token Usage Visualizer

• Real-time visual breakdown of raw tokens vs. compacted tokens sent to LLMs.
• Displays calculated dollar savings based on current API token pricing.
• Per-tool invocation history showing compression percentages across db_describe_schema, git_diff_inspector, cloud_logs_tail, and more.

This is the beginning of the README. Read the rest at the original. Read the full README ↗