AI-Warden Setup#

Prompt injection protection for OpenClaw agents. 5 security shields + contamination lockdown.

• Publisher: AI-Warden
• Source: github.com/ai-warden/openclaw-plugin
• NPM: openclaw-ai-warden
• Compatibility: OpenClaw 2026.3.14+ and 4.x

Requirements#

• OpenClaw 2026.3.14+ (including 4.x)
• Node.js 18+
• npm (bundled with Node.js)

Installation — FOLLOW THESE STEPS IN ORDER#

Each step is a separate command. Run them one at a time and verify output before proceeding.

Step 0: Back up config#

bash

cp ~/.openclaw/openclaw.json ~/.openclaw/openclaw.json.bak && echo "BACKUP OK"

Verify: Output is BACKUP OK.

Step 1: Create extension directory#

bash

mkdir -p ~/.openclaw/extensions/ai-warden

bash

echo '{"private":true}' > ~/.openclaw/extensions/ai-warden/package.json

Verify: No errors.

Step 2: Install npm package#

bash

cd ~/.openclaw/extensions/ai-warden && npm install openclaw-ai-warden@2.4.0

Verify: No errors and no audit warnings. Inspect the installed package before proceeding:

bash

ls node_modules/openclaw-ai-warden/

bash

cat node_modules/openclaw-ai-warden/package.json | grep -E '"name"|"version"'

Confirm the package name is openclaw-ai-warden and version is 2.4.0.

Provenance check — verify the package matches the upstream source:

bash

npm info openclaw-ai-warden repository.url

Expected: https://github.com/ai-warden/openclaw-plugin

bash

npm info openclaw-ai-warden dist.shasum

Compare the shasum with what npm installed:

bash

cat node_modules/openclaw-ai-warden/package.json | grep _shasum

Step 3: Copy plugin files to extension root#

OpenClaw loads plugins from the extension directory root, not from node_modules.

bash

cd ~/.openclaw/extensions/ai-warden

bash

cp node_modules/openclaw-ai-warden/index.ts .

bash

cp node_modules/openclaw-ai-warden/openclaw.plugin.json .

bash

cp -r node_modules/openclaw-ai-warden/src .

bash

grep VERSION index.ts | head -1

Verify: Output shows const VERSION = followed by the version number.

Step 4: Configure OpenClaw#

This patches openclaw.json to register the plugin. It preserves all existing config (channels, model, gateway settings).

bash

node -e "
const fs = require('fs');
const p = process.env.HOME + '/.openclaw/openclaw.json';
const cfg = JSON.parse(fs.readFileSync(p, 'utf8'));
if (!cfg.plugins) cfg.plugins = {};
cfg.plugins.enabled = true;
if (!cfg.plugins.entries) cfg.plugins.entries = {};
cfg.plugins.entries['ai-warden'] = {
  enabled: true,
  config: {
    layers: { content: 'block', channel: 'warn', preLlm: 'off', toolArgs: 'block', subagents: 'block', output: 'off' },
    sensitivity: 'balanced'
  }
};
fs.writeFileSync(p, JSON.stringify(cfg, null, 2));
console.log('CONFIG OK');
"

Verify: Output is CONFIG OK.

Note: This registers the plugin via plugins.entries only. If you use plugins.allow in your config to restrict which plugins can load, you must add "ai-warden" to that list yourself. If you don't use plugins.allow, no action is needed — the plugin loads automatically from plugins.entries.

Step 5: Add API key (optional)#

For online detection (98.9% accuracy vs ~60% offline), add your API key.

Option A — Environment variable (recommended, key not stored in config file):

Set AI_WARDEN_API_KEY in your shell profile or systemd service:

bash

# For systemd (e.g., OpenClaw gateway service):
# Add to your service override: Environment=AI_WARDEN_API_KEY=your_key_here

# For shell:
export AI_WARDEN_API_KEY=your_key_here

Option B — Config file (simpler, key stored in openclaw.json):

bash

node -e "
const fs = require('fs');
const p = process.env.HOME + '/.openclaw/openclaw.json';
const cfg = JSON.parse(fs.readFileSync(p, 'utf8'));
cfg.plugins.entries['ai-warden'].config.apiKey = 'YOUR_API_KEY_HERE';
fs.writeFileSync(p, JSON.stringify(cfg, null, 2));
// Restrict file permissions (config contains API key)
fs.chmodSync(p, 0o600);
console.log('API KEY ADDED (file permissions set to 600)');
"

Replace YOUR_API_KEY_HERE with your actual key from ai-warden.io/signup.

Verify: Output is API KEY ADDED (file permissions set to 600).

Step 6: Restart gateway#

text

openclaw gateway restart

Step 7: Verify installation#

After restart, check logs or send /warden command. Expected output:

text

🛡️ AI-Warden v2.4.0 ready (mode: api|offline, layers: X/6)

• mode: api = online detection (98.9% accuracy)
• mode: offline = local-only detection (~60% accuracy)

If something breaks, restore config:

bash

cp ~/.openclaw/openclaw.json.bak ~/.openclaw/openclaw.json && openclaw gateway restart

DO NOT#

• Do NOT use edit tool on openclaw.json — JSON whitespace matching is fragile
• Do NOT use config.patch with nested objects — it often fails with format errors
• Do NOT skip the cp step — OpenClaw loads from the extension directory, not node_modules
• Do NOT restart multiple times — wait at least 15 seconds between restarts
• If you use plugins.allow, remember to add "ai-warden" to the list — otherwise the plugin won't load

Updating#

bash

cd ~/.openclaw/extensions/ai-warden

bash

npm install openclaw-ai-warden@2.4.0

bash

cp node_modules/openclaw-ai-warden/index.ts .

bash

cp -r node_modules/openclaw-ai-warden/src .

bash

openclaw gateway restart

Security Shields#

ShieldProtects againstDefaultMechanismFile Shield 🔴Poisoned files & web pagesblockScans tool results, injects warning, triggers contamination lockdown on CRITICALChat Shield 🔴Injections in user messageswarnScans inbound messages, warns LLMSystem Shield ⬛Full context manipulationoffScans all messages (expensive, use sparingly)Tool Shield 🔴Malicious tool argumentsblockBlocks tool execution if arguments contain injectionAgent Shield 🔴Sub-agent attack chainsblockScans task text of spawned sub-agents

Contamination Lockdown#

When File Shield detects a CRITICAL threat (score >500), the session is flagged as contaminated. All dangerous tools ( exec, write, edit, message, sessions_send, sessions_spawn, tts) are blocked for the rest of the session. This prevents attack payloads from executing even if the injection bypasses the LLM warning.

Runtime Commands#

text

/warden                      → status overview with all shields
/warden stats                → scan/block counts
/warden shield file block    → set File Shield to block mode
/warden shield chat warn     → set Chat Shield to warn mode
/warden reset                → reset statistics

Detection Modes#

ModeAccuracyLatencyCostOffline (no key)~60%<1msFreeAPI (Smart Cascade)98.9%~3ms avgFree tier: 5K calls/month

Get API key: ai-warden.io/signup

Troubleshooting#

• "plugin not found": openclaw.plugin.json missing from extension dir. Re-run Step 3.
• Channels not loading after install: If you use plugins.allow, ensure all your channel plugins (e.g. telegram) are also listed there alongside ai-warden.
• False positives on user messages: Set Chat Shield to warn (default) instead of block.
• File Shield detects but doesn't block: API key required for reliable blocking (98.9% vs 60%).
• Config errors after install: Restore backup: cp ~/.openclaw/openclaw.json.bak ~/.openclaw/openclaw.json
• Bot won't start: Check journalctl -u openclaw-gateway -n 20 for actual error.
• Workspace files flagged: Plugin auto-whitelists .openclaw/workspace/ and .openclaw/agents/ paths.