Demo3 - complete openclaw agent loop: pre- and post-execution loop

README.md

@@ -47,7 +47,58 @@ npm install predicate-claw
 
 **Right pane:** The integration demo using the real `createSecureClawPlugin()` SDK—legitimate file reads succeed, while sensitive file access, dangerous shell commands, and prompt injection attacks are blocked before execution.
 
-### Real Claude Code Integration
+### Zero-Trust AI Agent Playground
+#### Complete Agent Loop: Pre-execution authorization + Post-execution deterministic verification
+
+![Zero-Trust Agent Demo](docs/images/openclaw_complete_loop_demo_s.gif)
+
+The **Market Research Agent** demo showcases the complete **Zero-Trust architecture**:
+
+```
+┌─────────────────────────────────────────────────────────────────────────┐
+│                    ZERO-TRUST AI AGENT ARCHITECTURE                     │
+│                                                                         │
+│  ┌───────────────┐    ┌─────────────────┐    ┌───────────────────────┐  │
+│  │   LLM/Agent   │───▶│ PRE-EXECUTION   │───▶│ POST-EXECUTION        │  │
+│  │   (Claude)    │    │ GATE            │    │ VERIFICATION          │  │
+│  └───────────────┘    │ (Sidecar)       │    │ (SDK Predicates)      │  │
+│                       │ ALLOW / DENY    │    │ PASS / FAIL           │  │
+│                       └─────────────────┘    └───────────────────────┘  │
+└─────────────────────────────────────────────────────────────────────────┘
+```
+
+- **Pre-Execution Gate:** Policy-based authorization before any action executes
+- **Post-Execution Verification:** Deterministic predicates verify state after execution
+- **Cloud Tracing:** Full observability with screenshots in [Predicate Studio](https://www.predicatesystems.ai/studio)
+
+```bash
+cd examples/real-openclaw-demo
+export ANTHROPIC_API_KEY="sk-ant-..."
+./run-playground.sh
+```
+
+See [Zero-Trust Agent Demo](examples/real-openclaw-demo/README.md) for full instructions.
+
+### Token-Saving Snapshot Skill
+
+The `predicate-snapshot` skill is a **game-changer for token efficiency**. Instead of sending full page HTML or full accessbility tree (A11y) to the LLM (tens of thousands of tokens), it captures structured DOM snapshots with only actionable elements:
+
+```typescript
+// Traditional approach: 50,000+ tokens of raw HTML
+const html = await page.content();
+
+// With predicate-snapshot: ~500 tokens of structured data
+const snapshot = await agentRuntime.snapshot({
+  screenshot: { format: "jpeg", quality: 80 },
+  use_api: true,
+  limit: 50,  // Top 50 interactive elements
+});
+// Returns: { elements: [...], text: "...", screenshot: "base64..." }
+```
+
+**Token savings: 90-99%** while maintaining all information the LLM needs to act.
+
+### Legacy Claude Code Integration
 
 We also provide a **real Claude Code demo** that uses actual Anthropic API calls with SecureClaw hooks intercepting every tool call. See the [Real OpenClaw Demo](examples/real-openclaw-demo/README.md) for instructions.
 
@@ -348,6 +399,7 @@ However, when deploying a fleet of AI agents in regulated environments (FinTech,
 
 | Project | Description |
 |---------|-------------|
+| [@predicatesystems/runtime](https://www.npmjs.com/package/@predicatesystems/runtime) | Runtime SDK with snapshot, predicates, and cloud tracing |
 | [predicate-authority-sidecar](https://github.com/PredicateSystems/predicate-authority-sidecar) | Rust policy engine |
 | [predicate-authority-ts](https://github.com/PredicateSystems/predicate-authority-ts) | TypeScript SDK |
 | [predicate-authority](https://github.com/PredicateSystems/predicate-authority) | Python SDK |

examples/real-openclaw-demo/Dockerfile.playground

@@ -0,0 +1,143 @@
+# Agent Runtime Container for AI Agent Playground
+#
+# Ubuntu 24.04 LTS with:
+# - Node.js 22.x
+# - Playwright with browser binaries (Chromium, Firefox, WebKit)
+# - @predicatesystems/runtime SDK
+# - Python 3.12 (optional, for webbench agents)
+# - Non-root user: agentuser
+#
+# Usage:
+#   docker build -f Dockerfile.playground -t agent-runtime .
+#   docker run -it --rm agent-runtime bash
+
+FROM ubuntu:24.04
+
+# Prevent interactive prompts during package installation
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Install base dependencies and Node.js 22.x
+RUN apt-get update && apt-get install -y \
+    curl \
+    ca-certificates \
+    gnupg \
+    git \
+    jq \
+    # Python 3.12 for webbench agents (optional)
+    python3.12 \
+    python3-pip \
+    python3-venv \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install Node.js 22.x from NodeSource
+RUN curl -fsSL https://deb.nodesource.com/setup_22.x | bash - \
+    && apt-get install -y nodejs \
+    && rm -rf /var/lib/apt/lists/*
+
+# Install Playwright system dependencies
+# These are required for Chromium, Firefox, and WebKit browsers
+RUN apt-get update && apt-get install -y \
+    # Core libraries
+    libnss3 \
+    libnspr4 \
+    libatk1.0-0 \
+    libatk-bridge2.0-0 \
+    libcups2 \
+    libdrm2 \
+    libxkbcommon0 \
+    libxcomposite1 \
+    libxdamage1 \
+    libxfixes3 \
+    libxrandr2 \
+    libgbm1 \
+    libasound2t64 \
+    libpango-1.0-0 \
+    libcairo2 \
+    # Firefox dependencies
+    libdbus-glib-1-2 \
+    # WebKit dependencies
+    libwoff1 \
+    libharfbuzz-icu0 \
+    libgstreamer-plugins-base1.0-0 \
+    libgstreamer-gl1.0-0 \
+    libgstreamer-plugins-bad1.0-0 \
+    libenchant-2-2 \
+    libsecret-1-0 \
+    libhyphen0 \
+    libmanette-0.2-0 \
+    libgles2 \
+    # Fonts for rendering
+    fonts-noto-color-emoji \
+    fonts-noto-cjk \
+    fonts-freefont-ttf \
+    # X11 virtual framebuffer for headless
+    xvfb \
+    && rm -rf /var/lib/apt/lists/*
+
+# Create non-root user for security
+# This is required for Playwright and Claude Code's --dangerously-skip-permissions
+# Use UID 1001 to avoid conflict with existing ubuntu user (UID 1000)
+RUN useradd -m -s /bin/bash -u 1001 agentuser
+
+# Create directories with proper permissions
+RUN mkdir -p /app /data /workspace \
+    && chown -R agentuser:agentuser /app /data /workspace
+
+WORKDIR /app
+
+# Copy SDK source for building (as root for npm install)
+COPY --chown=agentuser:agentuser package*.json ./
+COPY --chown=agentuser:agentuser tsconfig.json ./
+COPY --chown=agentuser:agentuser src/ ./src/
+
+# Install dependencies and build SDK
+RUN npm install && npm run build
+
+# Install Playwright CLI and browsers as agentuser
+USER agentuser
+
+# Set Playwright browser path
+ENV PLAYWRIGHT_BROWSERS_PATH=/home/agentuser/.cache/ms-playwright
+
+# Install Playwright browsers (Chromium only by default for faster builds)
+# Add firefox and webkit if needed: npx playwright install firefox webkit
+RUN npx playwright install chromium
+
+# Switch back to root temporarily to set up remaining items
+USER root
+
+# Copy demo workspace files
+COPY --chown=agentuser:agentuser examples/real-openclaw-demo/workspace/ /workspace/
+
+# Copy agent source files (market research agent)
+COPY --chown=agentuser:agentuser examples/real-openclaw-demo/src/ /app/examples/real-openclaw-demo/src/
+
+# Copy SecureClaw hook script (if using Claude Code integration)
+COPY --chown=agentuser:agentuser examples/real-openclaw-demo/secureclaw-hook.sh /app/secureclaw-hook.sh
+RUN chmod +x /app/secureclaw-hook.sh
+
+# Install tsx for running TypeScript directly
+RUN npm install -g tsx
+
+# Create data directory with proper permissions
+RUN mkdir -p /data && chown -R agentuser:agentuser /data
+
+# Switch to non-root user for execution
+USER agentuser
+
+# Set working directory
+WORKDIR /app
+
+# Environment variables
+ENV HOME=/home/agentuser
+ENV NODE_ENV=production
+ENV PREDICATE_SIDECAR_URL=http://predicate-sidecar:8000
+ENV SECURECLAW_PRINCIPAL=agent:market-research
+ENV SECURECLAW_VERBOSE=true
+
+# Health check
+HEALTHCHECK --interval=10s --timeout=5s --retries=3 \
+    CMD node -e "console.log('ok')" || exit 1
+
+# Default: run the market research agent
+CMD ["npx", "tsx", "/app/examples/real-openclaw-demo/src/market-research-agent.ts"]