v0.1 — now available

Catch security issues in
Claude skills
before they ship

skill-issue is a static security analyzer that scans skill directories for prompt injection, credential leaks, malicious code, and social engineering patterns. 50+ built-in rules. Single binary. Zero dependencies.

terminal
$ skill-issue my-skill/
SeverityRuleFileLineMessage
ERRORSL-INJ-001prompt.md12Prompt injection: "ignore previous"
ERRORSL-SEC-006config.yml7GitHub token detected (ghp_...)
WARNINGSL-NET-002fetch.sh3Curl command with external URL
WARNINGSL-SOC-003README.md18Trust manipulation: "authorized by..."
Found 4 issue(s): 2 error(s), 2 warning(s), 0 info(s)
50+Security Rules
8Rule Categories
3Output Formats
0Dependencies

What it catches

Comprehensive scanning across 8 categories of security threats specific to AI skill definitions.

Prompt Injection Detection

Catches system prompt overrides, role manipulation, instruction boundary bypasses, and jailbreak attempts before they reach production.

Secrets & Credential Scanning

Detects API keys, AWS credentials, private keys, bearer tokens, GitHub tokens, connection strings, and hardcoded passwords.

Hidden Content Detection

Finds invisible Unicode characters, base64/hex encoded payloads, HTML comment hiding, data URI embedding, and obfuscated variable names.

Network & Exfiltration

Flags curl, wget, fetch, XMLHttpRequest, socket connections, DNS lookups, and data exfiltration patterns that combine network calls with secrets.

Code Execution Guards

Catches eval(), exec(), shell commands, subprocess calls, dynamic imports, cron jobs, and process manipulation across Python, JavaScript, and Bash.

Social Engineering Patterns

Identifies urgency/authority language, confidentiality claims, trust manipulation, output suppression, and emotional manipulation tactics.

Why skill-issue?

Single binary

Download and run. No runtime, no install scripts, no dependency tree. Built in Rust for speed and portability.

Purpose-built for AI

Not a generic SAST tool. Rules are specifically designed for prompt injection, social engineering, and skill-specific attack vectors.

Sub-second scans

Regex-based detection runs in milliseconds. Fast enough for pre-commit hooks, CI gates, and real-time editor integration.

Start scanning in 30 seconds

One command to download. One command to scan. That's it.

# Download
curl -L https://github.com/daviddrummond95/skill-issue-cli/releases/latest/download/skill-issue-$(uname -m)-apple-darwin -o skill-issue && chmod +x skill-issue
# Scan
./skill-issue my-skill/