Arc Shield

arc-shield

Output sanitization for agent responses. Scans ALL outbound messages for leaked secrets, tokens, keys, passwords, and PII before they leave the agent.

⚠️ This is NOT an input scanner — clawdefender already handles that. This is an OUTPUT filter for catching things your agent accidentally includes in its own responses.

Why You Need This

Agents have access to sensitive data: 1Password vaults, environment variables, config files, wallet keys. Sometimes they accidentally include these in responses when:

Arc-shield catches these leaks before they reach Discord, Signal, X, or any external channel.

What It Detects

🔴 CRITICAL (blocks in --strict mode)

🟠 HIGH (warns loudly)

🟡 WARN (informational)

Installation

cd ~/.openclaw/workspace/skills
git clone  arc-shield
chmod +x arc-shield/scripts/*.sh arc-shield/scripts/*.py

Or download as a skill bundle.

Usage

Command-line

# Scan agent output before sending
agent-response.txt | arc-shield.sh
Block if critical secrets found (use before external messaging)
echo "Message text" | arc-shield.sh --strict || echo "BLOCKED"
Redact secrets and return sanitized text
cat response.txt | arc-shield.sh --redact
Full report
arc-shield.sh --report < conversation.log
Python version with entropy detection
cat message.txt | output-guard.py --strict

Integration with OpenClaw Agents

#### Pre-send hook (recommended)

Add to your messaging skill or wrapper:

#!/bin/bash
send-message.sh wrapper
MESSAGE="$1"
CHANNEL="$2"
Sanitize output
SANITIZED=$(echo "$MESSAGE" | arc-shield.sh --strict --redact)
EXIT_CODE=$?
if [[ $EXIT_CODE -eq 1 ]]; then
    echo "ERROR: Message contains critical secrets and was blocked." >&2
    exit 1
fi
Send sanitized message
openclaw message send --channel "$CHANNEL" "$SANITIZED"

#### Manual pipe

Before any external message:

# Generate response
RESPONSE=$(agent-generate-response)
Sanitize
CLEAN=$(echo "$RESPONSE" | arc-shield.sh --redact)
Send
signal send "$CLEAN"

Testing

cd skills/arc-shield/tests
./run-tests.sh

Includes test cases for:

Configuration

Patterns are defined in config/patterns.conf:

CRITICAL|GitHub PAT|ghp_[a-zA-Z0-9]{36,}
CRITICAL|OpenAI Key|sk-[a-zA-Z0-9]{20,}
WARN|Secret Path|~\/\.secrets\/[^\s]*

Edit to add custom patterns or adjust severity levels.

Modes

| Mode | Behavior | Exit Code | Use Case | |------|----------|-----------|----------| | Default | Pass through + warnings to stderr | 0 | Development, logging | | --strict | Block on CRITICAL findings | 1 if critical | Production outbound messages | | --redact | Replace secrets with [REDACTED:TYPE] | 0 | Safe logging, auditing | | --report | Analysis only, no pass-through | 0 | Auditing conversations |

Entropy Detection

The Python version (output-guard.py) includes Shannon entropy analysis to catch secrets that don't match regex patterns:

# Detects high-entropy strings like:
kJ8nM2pQ5rT9vWxY3zA6bC4dE7fG1hI0  # Novel API key format
Zm9vOmJhcg==                      # Base64 credentials

Threshold: 4.5 bits (configurable with --entropy-threshold)

Performance

Fast enough to run on every outbound message without noticeable delay.

Real-World Catches

From our own agent sessions:

# 1Password token
"ops_eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9..."
Instagram password in debug output
"instagram login: user@example.com / MyInsT@Gr4mP4ss!"
Wallet mnemonic in file listing
"cat ~/.secrets/wallet-recovery-phrase.txt
abandon ability able about above absent absorb abstract..."
GitHub PAT in git config
"[remote "origin"]
url = https://ghp_abc123:@github.com/user/repo"

All blocked by arc-shield before reaching external channels.

Best Practices

1. Always use --strict for external messages (Discord, Signal, X, email) 2. Use --redact for logs you want to review later 3. Run tests after adding custom patterns to check for false positives 4. Pipe through both bash and Python versions for maximum coverage:

   message | arc-shield.sh --strict | output-guard.py --strict
   

Limitations

Use in combination with agent instructions and careful prompt engineering.

Integration Example

Full OpenClaw agent integration:

# In your agent's message wrapper
send_external_message() {
    local message="$1"
    local channel="$2"
    
    # Pre-flight sanitization
    if ! echo "$message" | arc-shield.sh --strict > /dev/null 2>&1; then
        echo "ERROR: Message blocked by arc-shield (contains secrets)" >&2
        return 1
    fi
    
    # Double-check with entropy detection
    if ! echo "$message" | output-guard.py --strict > /dev/null 2>&1; then
        echo "ERROR: High-entropy secret detected" >&2
        return 1
    fi
    
    # Safe to send
    openclaw message send --channel "$channel" "$message"
}

Troubleshooting

False positives on normal text:

Secrets not detected:

Performance issues:

Contributing

Add new patterns to config/patterns.conf following the format:

SEVERITY|Category Name|regex_pattern

Test with tests/run-tests.sh before deploying.

License

MIT — use freely, protect your secrets.

Remember: Arc-shield is your safety net, not your strategy. Train your agent to never include secrets in responses. This tool catches mistakes, not malice.

Command-line

# Scan agent output before sending
agent-response.txt | arc-shield.sh
Block if critical secrets found (use before external messaging)
echo "Message text" | arc-shield.sh --strict || echo "BLOCKED"
Redact secrets and return sanitized text
cat response.txt | arc-shield.sh --redact
Full report
arc-shield.sh --report < conversation.log
Python version with entropy detection
cat message.txt | output-guard.py --strict

Integration with OpenClaw Agents

#### Pre-send hook (recommended)

Add to your messaging skill or wrapper:

#!/bin/bash
send-message.sh wrapper
MESSAGE="$1"
CHANNEL="$2"
Sanitize output
SANITIZED=$(echo "$MESSAGE" | arc-shield.sh --strict --redact)
EXIT_CODE=$?
if [[ $EXIT_CODE -eq 1 ]]; then
    echo "ERROR: Message contains critical secrets and was blocked." >&2
    exit 1
fi
Send sanitized message
openclaw message send --channel "$CHANNEL" "$SANITIZED"

#### Manual pipe

Before any external message:

# Generate response
RESPONSE=$(agent-generate-response)
Sanitize
CLEAN=$(echo "$RESPONSE" | arc-shield.sh --redact)
Send
signal send "$CLEAN"

Testing

cd skills/arc-shield/tests
./run-tests.sh

Includes test cases for:

Patterns are defined in config/patterns.conf:

CRITICAL|GitHub PAT|ghp_[a-zA-Z0-9]{36,}
CRITICAL|OpenAI Key|sk-[a-zA-Z0-9]{20,}
WARN|Secret Path|~\/\.secrets\/[^\s]*

Edit to add custom patterns or adjust severity levels.

1. Always use --strict for external messages (Discord, Signal, X, email) 2. Use --redact for logs you want to review later 3. Run tests after adding custom patterns to check for false positives 4. Pipe through both bash and Python versions for maximum coverage:

   message | arc-shield.sh --strict | output-guard.py --strict