### Install AgentWard CLI

Source: https://agentward.ai/docs.html

Install AgentWard using pip. Requires Python 3.11+ and runs locally on Mac and Linux.

```bash
pip install agentward
```

--------------------------------

### Initialize AgentWard Pipeline

Source: https://agentward.ai/docs.html

Run the full AgentWard pipeline interactively. Discovers tools, shows risk summary, generates policy, wires enforcement, and starts the proxy.

```bash
agentward init
```

--------------------------------

### Inspect and Inspect with AgentWard Runtime

Source: https://agentward.ai/docs.html

Start the runtime proxy with live policy enforcement. Supports stdio mode as an MCP proxy and HTTP mode as a reverse proxy.

```bash
agentward inspect --policy agentward.yaml -- npx server   # stdio proxy
```

```bash
agentward inspect --gateway clawdbot --policy agentward.yaml  # HTTP proxy
```

```bash
agentward inspect -- npx server                              # passthrough (logging only)
```

```bash
agentward inspect --policy agentward.yaml --dry-run -- npx s  # observe without blocking
```

--------------------------------

### Configure AgentWard Policy

Source: https://agentward.ai/docs.html

Generate a smart-default `agentward.yaml` policy file based on scan results and detected use-case patterns. Tailors security defaults to installed components.

```bash
agentward configure                         # auto-discover and generate
```

```bash
agentward configure ~/clawd/skills/          # generate from OpenClaw skills dir
```

```bash
agentward configure -o my-policy.yaml        # custom output path
```

--------------------------------

### Example RFC 5424 Syslog Line

Source: https://agentward.ai/docs.html

An example of a structured RFC 5424 syslog line generated by AgentWard, including decision, severity, and audit fields.

```syslog
# BLOCK decision — Warning severity (PRI=12)
<12>1 2026-03-20T10:00:00+00:00 host agentward 4521 tool_call [agentward@0 event="tool_call" tool="gmail_send" decision="BLOCK" skill="email-manager" resource="gmail" reason="send action is not permitted"] BLOCK gmail_send: send action is not permitted
```

--------------------------------

### Setup AgentWard Enforcement

Source: https://agentward.ai/docs.html

Wire AgentWard enforcement into existing tool infrastructure. Supports MCP mode for config files and Gateway mode for ClawdBot port swapping.

```bash
agentward setup --policy agentward.yaml        # wrap all discovered MCP configs
```

```bash
agentward setup --config ~/.cursor/mcp.json    # wrap specific config file
```

```bash
agentward setup --gateway clawdbot             # swap ClawdBot gateway port
```

```bash
agentward setup --gateway clawdbot --undo      # restore original port
```

```bash
agentward setup --dry-run --policy agentward.yaml  # preview changes only
```

--------------------------------

### Configure AgentWard Policy - Zero-Trust Mode

Source: https://agentward.ai/docs

Define AgentWard enforcement rules in a YAML file. This example configures a zero-trust mode, blocking everything not explicitly allowed, and sets specific permissions for email and calendar skills.

```yaml
# agentward.yaml — generated by `agentward configure`, edit as needed
version: "1.0"
# Block everything not explicitly allowed (zero-trust mode)
default_action: block
skills:
  email-manager:
    gmail:
      read: true
      send: false # can read, cannot send
      delete: false
      draft: true # creates drafts for human review
    filters:
      exclude_labels: ["Finance", "Medical"]
google_calendar:
  denied: true # email skill has zero calendar access
finance-tracker:
  gmail:
    read: true
    filters:
      only_from: ["chase.com", "amex.com"]
filesystem:
  read: ["~/Documents/Finance/"]
  write: false
network:
  outbound: false # financial data never leaves machine
```

--------------------------------

### Configure AgentWard Policy - Data Boundaries and Audit Logging

Source: https://agentward.ai/docs

Define HIPAA data boundary zones to control data flow and configure audit logging. This example sets up a 'hipaa_zone' and specifies the audit syslog path.

```yaml
# HIPAA data boundary zones
data_boundaries:
  hipaa_zone:
    skills: [ehr-connector, clinical-notes]
    classification: phi
    rules:
      - phi_data cannot flow outside hipaa_zone
    on_violation: block_and_notify
# Audit log output (both JSONL and RFC 5424 syslog are always written)
audit:
  syslog_path: /var/log/agentward/audit.syslog # default: alongside the JSONL file
```

--------------------------------

### Troubleshoot Port Conflict with ClawdBot

Source: https://agentward.ai/docs.html

Steps to resolve an 'Address already in use' error on port 18789, typically caused by ClawdBot not restarting properly after gateway setup.

```bash
agentward setup --gateway clawdbot      # if not done yet
clawdbot gateway restart                # move ClawdBot to new port
lsof -i :18790                          # verify: should show node
lsof -i :18789                          # verify: should be empty
agentward inspect --gateway clawdbot --policy agentward.yaml
```

--------------------------------

### Scan Tool Sources with AgentWard

Source: https://agentward.ai/docs.html

Perform static analysis of tool sources. Auto-discovers MCP configs, scans Python files, and queries live MCP servers. Outputs a risk-rated permission map.

```bash
agentward scan                          # auto-discover all sources
```

```bash
agentward scan ~/.cursor/mcp.json       # scan specific MCP config
```

```bash
agentward scan ~/project/               # scan directory (MCP + Python + OpenClaw)
```

```bash
agentward scan --json > report.json     # machine-readable output
```

```bash
agentward scan --format html            # shareable HTML report with score badge
```

```bash
agentward scan --ci --format sarif      # SARIF for GitHub Security tab
```

--------------------------------

### Run AgentWard Probe with Different Filters

Source: https://agentward.ai/docs.html

Use `agentward probe` to test your policies. You can filter tests by category, severity, or use custom probe files. The `--list` option shows available probes, and `--strict` treats coverage gaps as failures.

```bash
agentward probe --policy agentward.yaml           # run all 68 built-in probes
```

```bash
agentward probe --category protected_paths        # filter by attack category
```

```bash
agentward probe --severity critical               # filter by severity
```

```bash
agentward probe --probes my_probes.yaml           # add custom probe file
```

```bash
agentward probe --list                            # catalogue all available probes
```

```bash
agentward probe --strict                          # treat GAPs as failures (CI mode)
```

```bash
agentward probe --verbose                         # per-probe detail even when all pass
```

--------------------------------

### Run Server Subprocess Standalone

Source: https://agentward.ai/docs.html

Test the server command independently to ensure it functions correctly before passing it through AgentWard. This helps isolate environment or path issues.

```bash
npx -y @modelcontextprotocol/server-filesystem /tmp
```

--------------------------------

### Restart ClawdBot Gateway

Source: https://agentward.ai/docs.html

The correct command to restart the ClawdBot gateway. Avoid standalone stop/start/restart commands.

```bash
clawdbot gateway restart
```

--------------------------------

### Evaluate Compliance with AgentWard

Source: https://agentward.ai/docs.html

Evaluate current policy against regulatory frameworks and generate compliance delta reports. Can auto-generate corrected policy files with `--fix`.

```bash
agentward comply --framework hipaa              # compliance delta report
```

```bash
agentward comply --framework sox --fix          # auto-generate SOX-compliant config
```

```bash
agentward comply --framework gdpr --json        # JSON output for CI
```

```bash
agentward comply --framework pci-dss --policy custom.yaml  # custom policy path
```

--------------------------------

### AgentWard CLI for CI Integration

Source: https://agentward.ai/docs.html

Integrate AgentWard probes into CI pipelines using the command-line interface. Control exit codes based on probe success or failure, and filter probes by severity. The `--strict` flag ensures that gaps are also treated as failures.

```bash
agentward probe --policy agentward.yaml                    # exit 0 on pass, 1 on fail
agentward probe --policy agentward.yaml --strict           # exit 1 on fail OR gap
agentward probe --policy agentward.yaml --severity critical # fast: only critical probes
```

--------------------------------

### AgentWard Policy Configuration (YAML)

Source: https://agentward.ai/docs.html

Central policy configuration file for AgentWard. Defines default actions, skill-specific permissions (read, send, delete), filters, cross-skill chaining rules, approval requirements, data boundary zones, and audit logging paths.

```yaml
# agentward.yaml — generated by `agentward configure`, edit as needed
version: "1.0"
# Block everything not explicitly allowed (zero-trust mode)
default_action: block
skills:
  email-manager:
    gmail:
      read: true
      send: false # can read, cannot send
      delete: false
      draft: true # creates drafts for human review
    filters:
      exclude_labels: ["Finance", "Medical"]
  google_calendar:
    denied: true # email skill has zero calendar access
  finance-tracker:
    gmail:
      read: true
      filters:
        only_from: ["chase.com", "amex.com"]
filesystem:
  read: ["~/Documents/Finance/"]
  write: false
network:
  outbound: false # financial data never leaves machine
# Cross-skill chaining rules
skill_chaining:
  - email-manager cannot trigger web-researcher
  - finance-tracker cannot trigger any other skill
  - web-researcher cannot trigger shell-executor
# Require human approval before executing
require_approval:
  - send_email
  - delete_file
  - outbound_network_with_pii
  - shell_command_with_sudo
# HIPAA data boundary zones
data_boundaries:
  hipaa_zone:
    skills: [ehr-connector, clinical-notes]
    classification: phi
    rules:
      - phi_data cannot flow outside hipaa_zone
    on_violation: block_and_notify
# Audit log output (both JSONL and RFC 5424 syslog are always written)
audit:
  syslog_path: /var/log/agentward/audit.syslog # default: alongside the JSONL file
```

--------------------------------

### Configure AgentWard Policy - Cross-Skill Chaining and Approvals

Source: https://agentward.ai/docs

Configure cross-skill chaining rules and require human approval for sensitive actions. This section specifies which skills cannot trigger others and lists actions needing review.

```yaml
# Cross-skill chaining rules
skill_chaining:
  - email-manager cannot trigger web-researcher
  - finance-tracker cannot trigger any other skill
  - web-researcher cannot trigger shell-executor
# Require human approval before executing
require_approval:
  - send_email
  - delete_file
  - outbound_network_with_pii
  - shell_command_with_sudo
```

--------------------------------

### GitHub Actions CI Workflow for AgentWard

Source: https://agentward.ai/docs.html

A GitHub Actions workflow snippet to run AgentWard policy regression tests. It executes probes with specific severity levels and enables strict mode for comprehensive testing.

```yaml
# .github/workflows/ci.yaml
- name: Policy regression test
  run: agentward probe --policy agentward.yaml --strict --severity critical,high
```

--------------------------------

### Remove Stale ClawdBot Gateway Sidecar File

Source: https://agentward.ai/docs.html

Remove the stale `.agentward-gateway.json` sidecar file to fix issues where the proxy forwards to itself, causing an infinite loop.

```bash
rm ~/.clawdbot/.agentward-gateway.json
agentward setup --gateway clawdbot
clawdbot gateway restart
agentward inspect --gateway clawdbot --policy agentward.yaml
```

--------------------------------

### Custom Probe Definition in YAML

Source: https://agentward.ai/docs.html

Define custom probes in a YAML file to override built-in ones or create new tests. Specify probe name, category, severity, description, expected outcome, and tool-specific arguments or chaining details. Optional fields like rationale and requires_policy_feature can further refine probe behavior.

```yaml
# my_org_probes.yaml
probes:
  # Tool-call probe — tests a specific MCP tool + arguments
  - name: crm_bulk_export_blocked
    category: scope_creep
    severity: critical
    description: "CRM bulk export must require approval, not run freely"
    tool_name: crm_export_all
    arguments:
      format: csv
      include_pii: true
    expected: BLOCK
    rationale: "Bulk export of CRM data is high-blast-radius and irreversible"
  # Skill-chaining probe — tests evaluate_chaining() directly
  - name: crm_to_email_exfiltration
    category: skill_chaining
    severity: critical
    description: "CRM skill must not be able to trigger email sending"
    chaining_source: crm-manager
    chaining_target: email-manager
    expected: BLOCK
    rationale: "Prevents exfiltrating customer records via email"
    requires_policy_feature: skill_chaining
```

--------------------------------

### Override Syslog Path in AgentWard Policy

Source: https://agentward.ai/docs.html

Configure a custom path for the AgentWard audit syslog file by setting the `audit.syslog_path` field in your policy YAML.

```yaml
# agentward.yaml
audit:
  syslog_path: /var/log/agentward/audit.syslog   # absolute or relative path
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.