Security Audit Enhanced#

Advanced security audit framework for AI agents and system configurations. Combines knowledge-based guidance with automated scanning scripts.

Key Improvements Over Original#

FeatureOriginalEnhancedAutomationKnowledge onlyScripts + KnowledgeJSON Parsinggrep (unreliable)jq (proper)PlatformLinux onlymacOS + LinuxScoringNone0-100 scaleBaselineNoneTrack changesReportsText onlyJSON/HTML/Markdown

Quick Start#

Run Full Audit#

bash

python3 ~/.security-audit/scripts/audit.py --full

Quick Check (Critical Only)#

bash

python3 ~/.security-audit/scripts/audit.py --critical

Generate JSON Report#

bash

python3 ~/.security-audit/scripts/audit.py --json --output report.json

Compare with Baseline#

bash

python3 ~/.security-audit/scripts/audit.py --baseline ~/.security-audit/baseline.json

Security Scoring#

The audit produces a 0-100 security score:

ScoreRatingStatus90-100ExcellentMinimal risk70-89GoodMinor issues50-69FairNeeds attention30-49PoorSignificant risk0-29CriticalImmediate action required

The 13 Security Domains#

1. Gateway Exposure 🔴 Critical#

Risk: Unauthenticated network access to agent gateway.

Check:

bash

jq '.gateway.bind, .gateway.auth_token' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• bind: 127.0.0.1 or lan (not 0.0.0.0)
• auth_token: Set via env or config

Fix:

bash

export CLAWDBOT_GATEWAY_TOKEN="$(openssl rand -hex 32)"
# Or in config:
jq '.gateway.bind = "127.0.0.1"' ~/.clawdbot/clawdbot.json

---

2. DM Policy Configuration 🟠 High#

Risk: Any user can DM the bot and execute commands.

Check:

bash

jq '.channels | to_entries[] | select(.value.dmPolicy) | {channel: .key, policy: .value.dmPolicy}' ~/.clawdbot/clawdbot.json

Secure baseline:

• dmPolicy: allowlist or pairing (not open)

---

3. Group Access Control 🟠 High#

Risk: Anyone in groups can trigger bot commands.

Check:

bash

jq '.channels | to_entries[] | select(.value.groupPolicy) | {channel: .key, policy: .value.groupPolicy}' ~/.clawdbot/clawdbot.json

Secure baseline:

• groupPolicy: allowlist with explicit group IDs

---

4. Credentials Security 🔴 Critical#

Risk: Plaintext credentials with loose permissions.

Check:

bash

# Check credential files exist and permissions
ls -la ~/.clawdbot/credentials/ 2>/dev/null
# Check config permissions
stat -f "%Lp" ~/.clawdbot/clawdbot.json 2>/dev/null || stat -c "%a" ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• Directory permissions: 700
• File permissions: 600

Fix:

bash

chmod 700 ~/.clawdbot
chmod 600 ~/.clawdbot/clawdbot.json
chmod 600 ~/.clawdbot/credentials/*.json 2>/dev/null

---

5. Browser Control Exposure 🟠 High#

Risk: Remote browser control without authentication.

Check:

bash

jq '.browser.remoteControlUrl, .browser.remoteControlToken' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• remoteControlToken: Must be set if remote control enabled
• Use HTTPS for remote control URL

---

6. Network Binding 🟠 High#

Risk: Gateway exposed to public internet.

Check:

bash

jq '.gateway.bind, .gateway.mode, .gateway.tailscale' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• mode: local for development
• bind: 127.0.0.1 (localhost only)
• tailscale.mode: off unless intentionally used

---

7. Tool Access & Sandboxing 🟡 Medium#

Risk: Excessive tool permissions increase blast radius.

Check:

bash

jq '.restrict_tools, .mcp_tools, .workspaceAccess, .sandbox' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• restrict_tools: true
• workspaceAccess: ro or none
• sandbox: all for untrusted content

---

8. File Permissions 🟡 Medium#

Risk: Other users can read sensitive configs.

Check:

bash

# Check directory permission
stat -f "%Lp" ~/.clawdbot 2>/dev/null || stat -c "%a" ~/.clawdbot 2>/dev/null
# Check all JSON files
find ~/.clawdbot -name "*.json" -exec stat -f "%Lp %N" {} \; 2>/dev/null || find ~/.clawdbot -name "*.json" -exec stat -c "%a %n" {} \; 2>/dev/null

Secure baseline:

• ~/.clawdbot/: 700
• All .json files: 600

---

9. Plugin Trust 🟡 Medium#

Risk: Untrusted plugins can execute arbitrary code.

Check:

bash

jq '.plugins' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• Use explicit allowlist for plugins

---

10. Logging & Redaction 🟡 Medium#

Risk: Sensitive data leaks in logs.

Check:

bash

jq '.logging.redactSensitive' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• redactSensitive: tools or all

---

11. Prompt Injection Protection 🟡 Medium#

Risk: Untrusted content injects malicious prompts.

Check:

bash

jq '.wrap_untrusted_content, .untrusted_content_wrapper, .mentionGate' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline:

• wrap_untrusted_content: true
• mentionGate: true

---

12. Dangerous Command Blocking 🟡 Medium#

Risk: Destructive commands can be executed.

Check:

bash

jq '.blocked_commands' ~/.clawdbot/clawdbot.json 2>/dev/null

Secure baseline: Include patterns:

json

{
  "blocked_commands": [
    "rm -rf",
    "rm -rf /",
    "dd if=",
    "mkfs",
    ":(){ :|:& };:",
    "curl | bash",
    "wget | bash",
    "git push --force",
    "chmod 777",
    "> /dev/sda"
  ]
}

---

13. Secret Scanning 🟡 Medium#

Risk: Committed secrets in codebase.

Check:

bash

which detect-secrets 2>/dev/null && detect-secrets --version
ls -la .secrets.baseline 2>/dev/null

Secure baseline:

• detect-secrets installed
• .secrets.baseline exists and is current

---

Automated Scripts#

audit.py#

Main audit script with scoring and reporting.

bash

# Full audit with all checks
python3 scripts/audit.py --full

# Quick critical checks only
python3 scripts/audit.py --critical

# JSON output for CI/CD
python3 scripts/audit.py --json

# Compare with previous baseline
python3 scripts/audit.py --baseline baseline.json --diff

# Apply auto-fixes (safe only)
python3 scripts/audit.py --fix

check_permissions.py#

Standalone permission checker.

bash

python3 scripts/check_permissions.py --path ~/.clawdbot --fix

generate_report.py#

Generate formatted reports.

bash

# HTML report
python3 scripts/generate_report.py --format html --output report.html

# Markdown report
python3 scripts/generate_report.py --format markdown --output SECURITY.md

CI/CD Integration#

GitHub Actions#

yaml

name: Security Audit
on: [push, pull_request]
jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run Security Audit
        run: |
          pip install jq
          python3 scripts/audit.py --json --output audit-report.json
      - name: Upload Report
        uses: actions/upload-artifact@v4
        with:
          name: security-audit
          path: audit-report.json

Pre-commit Hook#

bash

#!/bin/bash
# .git/hooks/pre-commit
python3 scripts/audit.py --critical || {
  echo "Security audit failed. Fix issues before committing."
  exit 1
}

Baseline Management#

Create Baseline#

bash

python3 scripts/audit.py --save-baseline baseline.json

Compare with Baseline#

bash

python3 scripts/audit.py --baseline baseline.json

Update Baseline After Fixes#

bash

python3 scripts/audit.py --save-baseline baseline.json --force

Report Formats#

JSON Output#

json

{
  "timestamp": "2026-02-21T14:00:00Z",
  "score": 85,
  "rating": "Good",
  "summary": {
    "critical": 0,
    "high": 1,
    "medium": 2,
    "passed": 10
  },
  "findings": [
    {
      "domain": "dm-policy",
      "severity": "high",
      "finding": "DM policy is 'open'",
      "recommendation": "Set dmPolicy to 'allowlist'"
    }
  ]
}

HTML Report#

Generates styled HTML with:

• Score gauge visualization
• Severity breakdown chart
• Detailed findings table
• Remediation steps

Extending the Framework#

Add new checks by:

1. Create a new domain in audit.py:

python

def check_new_domain(config):
    issues = []
    # Your check logic
    if vulnerable:
        issues.append({
            'domain': 'new-domain',
            'severity': 'medium',
            'finding': 'Description',
            'recommendation': 'How to fix'
        })
    return issues

1. Add domain to checklist in SKILL.md

2. Run tests to validate

References#

• OWASP API Security: https://owasp.org/www-project-api-security/
• Clawdbot Security Docs: https://docs.clawd.bot/gateway/security
• Original Framework: https://github.com/TheSethRose/Clawdbot-Security-Check

---

Remember: Security is not a one-time check. Run audits regularly, track baselines, and respond to changes promptly.