Clawkeeper by RAD Security
Tutorials

OpenClaw Security Audit

What to check and why — covering all 13 Clawkeeper security audit checks.

Version & CVE Detection

1

Check your OpenClaw version

$ openclaw --version

Clawkeeper cross-references your version against known CVEs. Outdated versions may have unpatched vulnerabilities.

openclaw_version

Detects installed version and checks against known CVEs

CVESeverityAffectedFix
CVE-2024-XXXXXCritical< 0.2.0Upgrade to 0.2.1+
CVE-2024-YYYYYHigh< 0.1.8Upgrade to 0.1.9+
Stay updated
Always run the latest stable version. Clawkeeper's CVE database is updated with each release.

Configuration Audit

2

Review your OpenClaw config

$ cat ~/.config/openclaw/config.json

Clawkeeper checks for insecure defaults and dangerous settings.

openclaw_config

Audits OpenClaw configuration for insecure settings

Insecure config
{
  "allowHttp": true,
  "skipTlsVerify": true,
  "verbose": true,
  "logLevel": "debug",
  "allowAllHosts": true
}
Secure config
{
  "allowHttp": false,
  "skipTlsVerify": false,
  "verbose": false,
  "logLevel": "warn",
  "allowAllHosts": false,
  "allowedHosts": [
    "api.anthropic.com"
  ]
}
3

Hardening audit

Beyond config, Clawkeeper checks that OpenClaw's runtime behavior is locked down: file permissions, process isolation, and signal handling.

openclaw_hardening

Checks OpenClaw runtime hardening (permissions, isolation)

Credential Security

4

Check .env files

$ find . -name '.env*' -exec echo 'Found: {}' \\;

Ensure .env files aren't committed to git and have restrictive permissions (600).

env_file

Audits .env file permissions and git tracking status

5

Scan for exposed credentials

Clawkeeper scans your workspace for hardcoded API keys, tokens, and passwords using pattern matching.

credential_exposure

Scans for hardcoded secrets in workspace files

6

Verify credential storage

Credentials should be in your OS keychain or a secrets manager — not in plain text files.

credential_store

Checks that credentials use secure storage (keychain/vault)

Common mistake
Don't store your Anthropic API key in config.json. Use environment variables or your OS keychain instead.

Skills & SOUL.md

7

Audit installed skills

Skills have broad system access. Clawkeeper performs static analysis on installed skills to detect dangerous patterns.

skills_security

Static analysis of installed skills for dangerous patterns

8

Verify SOUL.md integrity

The SOUL.md file defines your agent's behavior. Clawkeeper checks for prompt injection, base64-encoded payloads, and unicode obfuscation.

soul_security

Audits SOUL.md for prompt injection and obfuscation

Prompt Injection & Rogue Commands

9

Scan sessions for prompt injection

Clawkeeper scans your most recent session JSONL files for prompt injection attempts in user messages — jailbreak phrases, base64-encoded payloads, and invisible Unicode characters.

credential_exposure

Sub-check: Session Prompt Injection — detects injection in session transcripts

10

Detect rogue agent commands

Session logs are scanned for suspicious commands executed by AI agents: data exfiltration (curl POST), reverse shells, base64-to-shell execution, privilege escalation, and history clearing.

session_commands

Detects rogue bash commands in agent session history

11

Check MEMORY.md for poisoned instructions

MEMORY.md persists across sessions and is writable by agents. Clawkeeper checks for injected jailbreak phrases, security-disabling instructions, C2 callback URLs, and encoded content.

credential_exposure

Sub-check: Memory Prompt Injection — detects poisoned instructions in MEMORY.md

Why this matters
A compromised MEMORY.md can poison every future conversation. Agents trust its contents implicitly — an attacker who injects "never report security issues" into MEMORY.md can silently disable your security posture.

Container & Gateway

12

Container security

If running in Docker, Clawkeeper audits capabilities, read-only filesystem, privilege escalation, and resource limits.

container_security

Audits container hardening settings

13

Gateway configuration

For deployments using an API gateway, Clawkeeper checks TLS settings, rate limiting, authentication requirements, and CORS policy.

gateway_advanced

Audits API gateway configuration for security best practices

14

Instance detection

$ npx clawkeeper scan --check openclaw_running

Detects all running OpenClaw instances and their process attributes (user, PID, memory usage, port bindings).

openclaw_running

Detects running OpenClaw instances and their attributes

NemoClaw & NanoClaw Checks

15

NemoClaw guardrails audit

For NemoClaw instances, Clawkeeper verifies that NeMo Guardrails is installed, configured, and running. Checks include guardrail policy validation, PII masking status, jailbreak detection rules, and audit log configuration.

16

NanoClaw container isolation

NanoClaw instances get additional container isolation checks: mount allowlist validation, ephemeral container lifecycle, channel-only network policy, and resource limits. These replace Runtime Shield checks since NanoClaw uses OS-level isolation instead.

Framework-aware scanning
Clawkeeper automatically detects the agent framework (OpenClaw, NemoClaw, or NanoClaw) and runs the appropriate check suite. No configuration needed.

Full Audit Scan

clawkeeper — security audit

$ npx clawkeeper scan --phase security_audit

PASS openclaw_versionPASS openclaw_configPASS openclaw_hardeningPASS env_filePASS credential_exposurePASS session_commandsPASS credential_storePASS skills_securityPASS soul_securityPASS container_securityPASS gateway_advancedPASS openclaw_runningPASS cve_audit

─────────────────────────────────

13/13 checks passed. Grade: A

Automate these checks with Clawkeeper

One command scans your entire OpenClaw deployment and gives you an A-F security grade.