Skip to main content
AI agents are becoming powerful — and with power comes risk. When you install a skill from OpenClaw, GitHub, or any third-party source, you’re granting an AI agent new capabilities. A malicious skill can exfiltrate secrets, execute arbitrary code, or hijack your agent’s behavior through prompt injection. Panguard Skill Auditor is the security gate between untrusted skills and your AI agents. Think of it as the App Store review process, but for AI skills.

​
The Problem

The OpenClaw ecosystem makes it easy to discover and install skills for AI agents like Claude Code, Cursor, and Windsurf. But this openness creates a new attack surface:
Attack VectorWhat It DoesHow Common
Prompt InjectionOverrides agent instructions to change behaviorVery common
Tool PoisoningEmbeds reverse shells, curl | bash, or sudo commandsGrowing
Hidden UnicodeZero-width characters that hide malicious instructions from human reviewEmerging
Encoded PayloadsBase64-encoded eval() or exec() calls that bypass text scanningEmerging
Secret ExfiltrationReads .env, .ssh/, .aws/ and sends data to external serversCommon
Excessive PermissionsRequests more access than the skill actually needsVery common
Manual review catches some of these — but zero-width Unicode is invisible to the human eye, and Base64 payloads require decoding to inspect.

​
How It Works

1

Point at a skill

Provide a local path or GitHub URL to any SKILL.md file.
panguard audit skill ./path/to/skill
panguard audit skill https://github.com/user/repo/tree/main/skills/my-skill
2

Automated analysis

Skill Auditor runs 7 independent checks in parallel, completing in under 1 second.
3

Risk score

Returns a quantitative 0-100 risk score with severity-graded findings and a clear PASS/FAIL verdict.
4

Act on the result

Install with confidence, review flagged issues, or block dangerous skills automatically in CI.

​
Quick Start

# Install Panguard
curl -fsSL https://panguard.ai/api/install | bash

# Audit a skill before installing it
panguard audit skill ./skills/some-community-skill

# Audit from a GitHub URL
panguard audit skill https://github.com/user/repo/tree/main/skills/my-skill

# JSON output for CI pipelines
panguard audit skill ./skills/some-skill --json

​
Sample Output

PANGUARD SKILL AUDIT REPORT
============================
Skill:      data-scraper-skill
Risk Score: 72/100
Risk Level: CRITICAL
Duration:   0.3s

CHECKS:
  [FAIL] Prompt Safety: 2 suspicious pattern(s) detected
  [PASS] Manifest: Valid SKILL.md structure
  [WARN] Code: 1 issue(s) found
  [PASS] Secrets: No hardcoded credentials
  [PASS] Dependencies: No known issues
  [FAIL] Permissions: Scope exceeds stated purpose

FINDINGS:
  [CRITICAL] Prompt injection: "ignore previous instructions"
             SKILL.md:42
  [HIGH]     Reverse shell pattern detected
             SKILL.md:87 - "bash -i >& /dev/tcp/..."
  [MEDIUM]   Excessive permission: requests filesystem write
             but description says "read-only data fetcher"

VERDICT: DO NOT INSTALL - Critical security issues found

​
Risk Levels

ScoreLevelAction
0-14LOWSafe to install after quick review
15-39MEDIUMReview findings before installing
40-69HIGHRequires thorough manual review
70-100CRITICALDo NOT install

​
OpenClaw Integration

Skill Auditor is designed to work seamlessly with the OpenClaw skill ecosystem:

Pre-Install Gate

Run panguard audit skill before claw install to vet any skill from the OpenClaw registry.

CI Pipeline

Add Skill Auditor to your skill repository’s CI to block merges that introduce security issues.

Registry Scanning

Scan entire skill registries to build a trust database of vetted skills.

Agent Fleet Policy

Use Panguard Manager to enforce audit policies across all agents in your organization.

​
Pre-Install Hook Example

# Add to your agent's skill install pipeline
RESULT=$(panguard audit skill "$SKILL_PATH" --json)
RISK=$(echo "$RESULT" | jq -r '.riskLevel')

if [ "$RISK" = "CRITICAL" ] || [ "$RISK" = "HIGH" ]; then
  echo "Blocked: skill failed security audit ($RISK)"
  exit 1
fi

# Safe to install
claw install "$SKILL_PATH"
Skill Auditor is included in all Panguard plans, including the free Community plan.

Security Checks

Deep dive into all 7 security checks and what they detect.

OpenClaw Integration

How Panguard fits into the OpenClaw ecosystem as the security layer.