DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to red-team a local model for prompt injection with Dodoguard

Run adversarial probes against a model hosted with Ollama to find jailbreaks and guardrail bypasses.

Short answer: Install and start Ollama on your machine and pull the model you want to test; then clone the repo and run npm ci && npm run build from the Dodoguard folder as in the quick start; open the examples/redteam-ollama folder in the repo and read its config. All steps.

Authorization notice. Dodoguard is dual-use. Only use it on software, sites and systems you own or have explicit permission to test.

Who it is forA security-minded engineer who wants to test a locally hosted model without sending prompts to a cloud provider.

The problemSystem prompts and guardrails can be bypassed, and teams need to find that before users do.

The outcomeA list of red-team findings for the local model, with evidence you can keep private.

What you need first

  • Node.js 22.12 or newer for the CLI
  • An OpenAI API key for the OpenAI example, or a local Ollama setup for the local red-team example
  • Docker with Docker Compose for the web console
  • Access to the Dify, Coze, FastGPT, or n8n app you want to check, if you run platform checks

Steps

  1. Install and start Ollama on your machine and pull the model you want to test
  2. Clone the repo and run npm ci && npm run build from the Dodoguard folder as in the quick start
  3. Open the examples/redteam-ollama folder in the repo and read its config
  4. Point the example config at your local Ollama model and the system prompt you want to test (inferred, check the README)
  5. Run the red-team example with the Dodoguard CLI from the examples/redteam-ollama folder and review the findings

Prompt for your agent

In the dodoguard repo, use the examples/redteam-ollama config to red-team a local Ollama model named <model-name>. Return the list of findings with severity and evidence. Verify that each finding links to the prompt that caused it.

Caveat. The README only links to the Ollama example without listing its exact commands, so confirm the run steps in that folder before you rely on them.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for dodoguardai/dodoguard and "Red-team a local model for prompt injection"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer