🦀 Pincer

Security scanner for OpenClaw — one script, zero dependencies.

Pincer audits your OpenClaw/Clawdbot/Moltbot installation for security misconfigurations and common vulnerabilities. Pure bash, runs anywhere, fixes what it finds.

Why?

• Forbes (Jan 2026): "Hundreds of Moltbot servers found exposed on the open internet"
• Cisco AI Defense: Documented prompt injection and malicious skill vectors
• CVE-2026-21636: Node.js permission model bypass affects OpenClaw setups

The problem: Most OpenClaw installations run with default settings — gateway on 0.0.0.0, no auth, no command allowlist, plaintext API keys. Pincer finds and fixes these issues in seconds.

Install

Quick scan (no install):

curl -sL https://raw.githubusercontent.com/masbindev/pincer/main/pincer.sh | bash

Install locally (recommended — enables fix command):

curl -sL https://raw.githubusercontent.com/masbindev/pincer/main/pincer.sh -o pincer.sh
chmod +x pincer.sh
./pincer.sh scan

Or clone the repo:

git clone https://github.com/masbindev/pincer.git
cd pincer && ./pincer.sh scan

Usage

# Full security audit
./pincer.sh scan

# JSON output (for CI/CD)
./pincer.sh scan --json

# Auto-fix critical issues
./pincer.sh fix

# Fix without confirmation prompt
./pincer.sh fix --yes

Sample Output

🦀 Pincer Security Report
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  Score: 65/100 (D)

🔴 CRITICAL
  ├─ Gateway exposed on 0.0.0.0 (should be 127.0.0.1)
  ├─ No shell command allowlist (safeBins not configured)
  └─ Node.js 22.12.0 may be vulnerable to CVE-2026-21636

🟡 WARNING
  ├─ HTTPS/TLS not configured
  ├─ Sensitive directories accessible (~/.ssh ~/.gnupg)
  └─ No rate limiting configured

🟢 PASSED
  ├─ API keys not found in plaintext configs
  ├─ Config file permissions OK
  ├─ Gateway auth token set and not using defaults
  └─ Control UI auth bypass disabled

💡 RECOMMENDATIONS
  ├─ Run: cisco-ai-defense/skill-scanner to audit your skills
  └─ See: https://github.com/masbindev/pincer for fix guides

Run ./pincer.sh fix to auto-fix critical issues.

Security Checks (12)

#	Check	Severity	What it detects
1	Gateway Binding	CRITICAL	Gateway listening on 0.0.0.0 instead of 127.0.0.1
2	API Keys Exposure	CRITICAL	Plaintext API keys/tokens in config files (not .env)
3	File Permissions	CRITICAL	World-readable config files (should be 600)
4	HTTPS/TLS	WARNING	No TLS/HTTPS configuration detected
5	Shell Command Allowlist	CRITICAL	Missing tools.exec.safeBins — agent can run any command
6	Sensitive Directories	WARNING	Agent can access ~/.ssh, ~/.gnupg, ~/.aws, /etc/shadow
7	Webhook Auth	WARNING	Webhook endpoints without authentication
8	Sandbox Isolation	WARNING	Not running in Docker/sandbox
9	Default/Weak Credentials	CRITICAL	Default tokens, undefined token bug, missing gateway auth
10	Rate Limiting	WARNING	No rate limiting configured
11	Node.js Version	CRITICAL	CVE-2026-21636 permission model bypass
12	Control UI Auth	CRITICAL	Auth bypass flag enabled

Fix Command

pincer.sh fix auto-remediates critical issues:

• Rebinds gateway to 127.0.0.1
• Sets config file permissions to 600
• Adds safeBins allowlist with sensible defaults
• Disables Control UI auth bypass
• Creates timestamped backup before any changes

How Pincer Differs

	Pincer	ClawShield (kappa9999)	Cisco Skill Scanner
Focus	Infra & config hardening	Audit + exposure + lockfile	Skill content scanning
Language	Pure bash	Go + Python	Python
Dependencies	Zero (coreutils only)	Go runtime, Python	Python, pip
Install	One curl command	Build from source	pip install
Auto-fix	✅ Yes	Partial	No
Complements	—	Overlaps	✅ Use together!

Recommended combo: Run Pincer for infrastructure hardening, then cisco-ai-defense/skill-scanner to audit skill contents for prompt injection.

Requirements

• Bash 4+
• Standard coreutils (grep, awk, stat, sed, ss)
• Works on any Linux distro, macOS with GNU coreutils

Contributing

PRs welcome! Please:

1. Keep it pure bash — no external dependencies
2. Run shellcheck pincer.sh before submitting
3. Add tests for new checks

License

MIT — see LICENSE