feat: add agent-governance-toolkit template

templates/config.json

@@ -21,6 +21,28 @@
     },
     "tags": ["AI Agents", "Developer Tools", "Automation"]
   },
+  {
+    "id": "agent-governance-toolkit",
+    "name": "microsoft/agent-governance-toolkit",
+    "description": "AI Agent Governance Toolkit — Policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering for autonomous AI agents. Covers 10/10 OWASP Agentic Top 10.",
+    "repo": "https://github.com/Phala-Network/phala-cloud/tree/main/templates/prebuilt/agent-governance-toolkit",
+    "author": "microsoft",
+    "icon": "agent-governance-toolkit.svg",
+    "envs": [
+      {
+        "key": "AGT_PACKAGE_VERSION",
+        "required": false,
+        "default": "3.7.0",
+        "description": "Published agent-governance-toolkit Python package version installed by the verifier at container startup."
+      }
+    ],
+    "defaultResource": {
+      "vCPU": 1,
+      "memory": 2048,
+      "diskSize": 20
+    },
+    "tags": ["AI Agents", "Developer Tools"]
+  },
   {
     "id": "VibeVM",
     "name": "VibeVM",

templates/prebuilt/agent-governance-toolkit/README.md

@@ -0,0 +1,128 @@
+# microsoft/agent-governance-toolkit
+
+Deploy a CPU-safe Agent Governance Toolkit package verifier on Phala Cloud.
+
+## Metadata
+
+- Template id: `agent-governance-toolkit`
+- Display name: `microsoft/agent-governance-toolkit`
+- Category: AI Agents & Developer Tools
+- Upstream repository: https://github.com/microsoft/agent-governance-toolkit
+- Upstream documentation: https://microsoft.github.io/agent-governance-toolkit/
+- Python package: `agent-governance-toolkit[full]`
+- Default package version: `3.7.0`
+- Icon source: upstream project icon, `docs/assets/agent-governance-toolkit.svg`
+- Upstream author: Microsoft, via the `microsoft/agent-governance-toolkit` GitHub repository
+
+## Overview
+
+Agent Governance Toolkit is Microsoft's public-preview toolkit for policy enforcement, zero-trust identity, execution sandboxing, auditability, compliance verification, and reliability controls for autonomous AI agents. The upstream README describes a Python-first quickstart with `pip install agent-governance-toolkit[full]`, `govern(...)` wrappers around tools, and CLI checks such as `agt verify` for OWASP Agentic Security Initiative coverage.
+
+This Phala Cloud template is a no-credential verifier, not a production agent service. It installs the real published Python package, imports `agentmesh.governance`, `agent_os.policies`, and `agent_compliance.cli.agt`, then exposes a deterministic HTTP smoke API.
+
+The `/demo` endpoint uses the real AGT `govern()` wrapper around a local Python tool, evaluates a YAML policy, allows a safe `read_file` action, blocks a destructive `delete_file` action, and returns compact audit hash-chain metadata. It does not call an LLM provider, download model weights, open browser authentication flows, require GPU access, or use external credentials.
+
+The upstream repository also includes a `docker-compose.yml`, but it is a development/test compose file with a local build context and bind-mounted workspace. This template avoids that shape so it is deployable on Phala Cloud.
+
+## Services
+
+- `app`: Internal Python HTTP server on port `8000`. At startup it installs `agent-governance-toolkit[full]` with `uv pip` and runs the deterministic verifier.
+- `proxy`: Caddy reverse proxy listening on public port `8080` and forwarding to `app:8000`.
+
+## Environment Variables
+
+No credentials are required.
+
+| Variable | Required | Default | Description |
+| --- | --- | --- | --- |
+| `AGT_PACKAGE_VERSION` | No | `3.7.0` | Published `agent-governance-toolkit` version installed by the verifier at container startup. |
+
+## Deploy On Phala Cloud
+
+1. Deploy the `agent-governance-toolkit` template.
+2. Keep the default CPU-only resources for this verifier.
+3. Leave `AGT_PACKAGE_VERSION` at `3.7.0` unless you intentionally want to test another published package version.
+4. Open `https://<your-app-domain>/healthz` after startup completes.
+
+The first startup downloads Python wheels from PyPI. The verifier path is local and deterministic after dependencies are installed.
+
+## Endpoints
+
+- `GET /healthz`: Reports package version, import checks, expected symbols, Python/runtime metadata, and no-LLM flags.
+- `GET /demo`: Runs the deterministic AGT `govern()` policy enforcement demo and returns allow, deny, and audit integrity results.
+- `GET /v1/models`: Returns an OpenAI-style metadata list for the local policy verifier. It is not a hosted model endpoint.
+- `GET /upstream`: Returns upstream repository, documentation, package, icon attribution, runtime shape, and endpoint metadata.
+- `GET /`: Returns a compact service index.
+
+Example:
+
+```bash
+curl -fsS https://<your-app-domain>/healthz
+curl -fsS https://<your-app-domain>/demo
+curl -fsS https://<your-app-domain>/v1/models
+curl -fsS https://<your-app-domain>/upstream
+```
+
+Expected `/demo` fields include:
+
+```json
+{
+  "ok": true,
+  "llm_provider_calls": false,
+  "model_downloaded": false,
+  "model_loaded": false,
+  "demo": {
+    "denied_call": {
+      "blocked": true,
+      "matched_rule": "block-destructive-actions"
+    },
+    "audit": {
+      "entry_count": 2,
+      "integrity_ok": true
+    }
+  }
+}
+```
+
+## Local Smoke Verification
+
+Run from the parent worktree:
+
+```bash
+docker compose -f sdks/templates/prebuilt/agent-governance-toolkit/docker-compose.yml config >/tmp/agt-compose.out
+docker compose -f sdks/templates/prebuilt/agent-governance-toolkit/docker-compose.yml up -d
+curl -fsS http://localhost:8080/healthz
+curl -fsS http://localhost:8080/demo
+curl -fsS http://localhost:8080/v1/models
+curl -fsS http://localhost:8080/upstream
+docker compose -f sdks/templates/prebuilt/agent-governance-toolkit/docker-compose.yml down
+```
+
+If local port `8080` is already in use, temporarily change only the host side of the proxy mapping, for example `18080:80`, then use `http://localhost:18080/healthz`.
+
+Template validation commands from the parent worktree:
+
+```bash
+python sdks/templates/validate.py
+git -C sdks diff --check origin/main...HEAD
+docker compose -f sdks/templates/prebuilt/agent-governance-toolkit/docker-compose.yml config >/dev/null
+```
+
+## Production Notes
+
+- This template verifies package installation and core governance primitives. It does not run a full production agent, sidecar fleet, dashboard, or managed policy service.
+- Add authentication, authorization, rate limiting, request logging, and network restrictions before exposing governance results for private workloads.
+- Real deployments should wrap actual framework tools or agent actions with AGT policy checks, configure identity/trust material, and store audit logs in durable infrastructure.
+- Do not put provider API keys, bearer tokens, private keys, OTPs, or passwords in the compose file. Use Phala Cloud secret handling for production credentials.
+- The default verifier is unauthenticated and should remain a smoke endpoint unless adapted behind an authenticated gateway.
+- The demo does not invoke model providers, download model weights, request GPU access, use privileged mode, mount host paths, use `env_file`, access the Docker socket, or use host networking.
+- The upstream project is public preview. Pin `AGT_PACKAGE_VERSION` for reproducible deployments and test policy behavior before changing versions.
+
+## Upstream Attribution
+
+This template installs and imports the real Agent Governance Toolkit Python package from Microsoft:
+
+- Repository: https://github.com/microsoft/agent-governance-toolkit
+- Documentation: https://microsoft.github.io/agent-governance-toolkit/
+- Package index: https://pypi.org/project/agent-governance-toolkit/
+- Icon: https://github.com/microsoft/agent-governance-toolkit/blob/main/docs/assets/agent-governance-toolkit.svg