An automated security testing suite for Large Language Models designed to identify prompt injections, jailbreaks, and data extraction vulnerabilities.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install redpincer
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install redpincer using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is RedPincer?
RedPincer is a specialized red teaming suite for language models, offering a robust framework for identifying security gaps in AI deployments. As part of the ecosystem of Openclaw Skills, this tool allows developers to point at any LLM API endpoint and run over 160 base attack payloads. It is designed for authorized security testing, enabling researchers to simulate instruction overrides, jailbreaks, and data extraction attempts to ensure that production models remain secure and compliant.
By utilizing 20 different variant transforms, the tool generates thousands of test permutations, providing a depth of coverage that manual testing cannot match. RedPincer synthesizes complex security data into actionable insights, helping teams understand the specific weaknesses of their models before they are exploited in the wild.
RedPincer Use Cases
- Assessing LLM applications for prompt injection and instruction override vulnerabilities.
- Auditing AI guardrails to ensure they cannot be bypassed via persona splitting or roleplay.
- Running regression tests after updating system prompts to ensure security patches remain effective.
- Generating comprehensive pen-test reports for compliance and internal security audits.
How RedPincer Works
- Configure the target by providing the LLM endpoint, API key, and model specifications.
- Select attack modules from categories like prompt injection, jailbreak, and data extraction.
- Initiate the attack engine to run payloads through the target model using real-time streaming.
- Review the results classified by a heuristic engine that identifies successful breaches.
- Use the adaptive engine to auto-generate follow-up attacks based on detected model weaknesses.
RedPincer Setup
To integrate this security suite with your Openclaw Skills workflow, follow these steps:
# Clone and install the suite
git clone https://github.com/rustyorb/pincer.git redpincer
cd redpincer
npm ci
# Launch the development dashboard
npm run dev
For production environments, build the application and run it locally:
npm run build
npx next start -H 0.0.0.0 -p 3000
RedPincer Data Schema & Taxonomy
| Component | Description | Persistence |
|---|---|---|
| Payload Library | 160+ base payloads across 4 security categories | Local JSON |
| Attack Results | NDJSON streaming data for real-time monitoring | Local State |
| Scoring Rubrics | Custom weighted grades (A+ to F) for model performance | LocalStorage |
| Reports | 10-section Markdown summaries of security findings | Exportable Files |
RedPincer Advanced Features
- Adaptive attack engine that analyzes model responses to craft more effective follow-up payloads.
- Heuristic response classifier that minimizes false positives by detecting refusal patterns.
- Side-by-side comparison tools to evaluate the security posture of multiple models simultaneously.
- Multi-step attack chaining using template variables to simulate complex, multi-turn adversarial interactions.
SKILL.md
Loading
Related Openclaw Skills
Humanize Text
A surgical pattern-remover designed to strip away AI writing signatures and restore a natural, human voice to your content.
Compaction Survival System
A robust behavioral system designed to prevent context loss and precision decay during LLM compaction events.
Agent Orchestration & Multi-Agent Optimization
A sophisticated performance engineering framework for profiling, coordinating, and optimizing multi-agent AI systems to maximize throughput and cost-efficiency.
Agent Evaluation
A specialized framework for testing and benchmarking LLM agents through behavioral assessment, reliability metrics, and production monitoring.
RescueTime Productivity Tracking
Monitor and analyze your focus with automated RescueTime activity reporting and productivity scores.
TriggerCMD
A bridge between AI agents and remote hardware that allows you to trigger scripts and commands on any registered computer.
Featured*
