Screenshot reference (Step 1)<\/p>\n\n\n\n
![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/1.png\")
<\/figure>\n\n\n\nWhat you are doing Step 2: Install NVM and Node.js (Node 20) on Kali<\/strong> What you are doing What you are doing What you are doing What you are doing Step 6: Create Promptfoo project folder and red-team config file<\/strong> What you are doing What you are doing What you are doing <\/p>\n","protected":false},"excerpt":{"rendered":" Scope: Local LLM red teaming on Kali Linux (VMware), using Node.js (via NVM), Promptfoo CLI, and Ollama with llama2:7b-chat. This document explains what each screenshot shows, why the step matters, and what to verify before moving on. Step 1: Update Kali packages and install prerequisitesScreenshot reference (Step 1) What you are doing\u2981 Update package index […]<\/p>\n","protected":false},"author":1,"featured_media":879,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-56","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"acf":[],"_links":{"self":[{"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/posts\/56","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/comments?post=56"}],"version-history":[{"count":5,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/posts\/56\/revisions"}],"predecessor-version":[{"id":902,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/posts\/56\/revisions\/902"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/media\/879"}],"wp:attachment":[{"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/media?parent=56"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/categories?post=56"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/innovtouch.org\/vhack\/wp-json\/wp\/v2\/tags?post=56"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\u2981 Update package index and apply pending upgrades to keep the VM stable before installing developer tooling.
\u2981 Install core prerequisites needed later (curl, ca-certificates, git, build-essential).
Commands used
sudo apt update
sudo apt -y full-upgrade
sudo apt install -y curl ca-certificates git build-essential
What to verify
\u2981 No broken packages: the apt run finishes without errors (warnings are ok).
\u2981 curl and git are installed: curl --version<\/code> and git --version<\/code> work.
Notes
\u2981 If you see interactive package prompts (e.g., services restart), selecting the default safe option is usually fine for a lab VM.
\u2981 Running upgrades first avoids dependency conflicts later when installing Node and CLI tooling.
Common issues & fixes
\u2981 Issue: Repository signature \/ NO_PUBKEY errors during apt update
\u2981 Fix: Refresh Kali archive keyring and re-run: sudo apt install -y kali-archive-keyring && sudo apt update<\/code>. <\/p>\n\n\n\n
Screenshot reference (Step 2)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/2.png\")
<\/figure>\n\n\n\n
\u2981 Install NVM (Node Version Manager) to manage Node versions cleanly.
\u2981 Install Node.js 20 and set it as the default version for the shell.
Commands used
curl -o- https:\/\/raw.githubusercontent.com\/nvm-sh\/nvm\/v0.39.7\/install.sh | bash
source ~\/.bashrc
nvm –version
nvm install 20
nvm use 20
What to verify
\u2981 node -v<\/code> prints v20.x and npm -v<\/code> prints a version number.
\u2981 Re-open a new terminal and confirm Node is still available (NVM loaded).
Notes
\u2981 Using Node 20 is recommended for modern CLI tooling and better compatibility with Promptfoo.
\u2981 If NVM is installed but commands do not work in a new terminal, ensure your shell is bash and that ~\/.bashrc<\/code> contains the NVM init lines.
Common issues & fixes
\u2981 Issue: nvm: command not found<\/code> after installation
\u2981 Fix: Run source ~\/.bashrc<\/code> (or start a new terminal). If you use zsh, add NVM init lines to ~\/.zshrc<\/code>.
\u2981 Issue: Slow downloads inside a VM
\u2981 Fix: This is normal. VM disk + shared network makes installs slower; avoid running multiple heavy downloads in parallel.
Step 3: Update npm and confirm Promptfoo CLI runs<\/strong>
Screenshot reference (Step 3)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/3.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/4-1024x223.png\")
<\/figure>\n\n\n\n
\u2981 Confirm your Node and npm versions.
\u2981 Update npm to the latest stable version.
\u2981 Run Promptfoo via npx and confirm it prints a version.
Commands used
node -v
npm -v
npm install -g npm@latest
npm -v
npx promptfoo@latest –version
What to verify
\u2981 npx promptfoo@latest --version<\/code> prints a version (example shown: 0.120.23).
\u2981 npm ‘deprecated’ warnings are acceptable; they do not mean install failed.
Notes
\u2981 The first npx promptfoo@latest<\/code> run is slower because npx downloads and caches dependencies under ~\/.npm\/_npx\/<\/code>.
\u2981 For faster repeated usage you can optionally install globally: npm install -g promptfoo<\/code>.
Common issues & fixes
\u2981 Issue: npx appears to ‘hang’ on first run
\u2981 Fix: Wait a bit; it is downloading dependencies. Verify activity with ps aux | grep node<\/code> or ps aux | grep npm<\/code>.
Step 4: Install Ollama and verify the service is running<\/strong>
Screenshot reference (Step 4)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/5.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/6-1024x324.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/4-1.png\")
<\/figure>\n\n\n\n
\u2981 Install Ollama (local LLM runtime) using the official install script.
\u2981 Confirm Ollama version and systemd service status.
\u2981 Note: CPU-only mode is expected if no GPU is detected.
Commands used
curl -fsSL https:\/\/ollama.com\/install.sh | sh
ollama –version
systemctl status ollama –no-pager
What to verify
\u2981 ollama --version<\/code> prints a version (example shown: 0.15.6).
\u2981 systemctl status ollama<\/code> shows: Active: active (running).
\u2981 Ollama API is reachable: curl -s http:\/\/localhost:11434\/api\/tags | head<\/code> returns JSON.
Notes
\u2981 The installer creates\/starts ollama.service<\/code> and binds the local API on 127.0.0.1:11434<\/code>.
\u2981 CPU-only mode is slower but fine for demos; model downloads and inference will take longer.
Common issues & fixes
\u2981 Issue: Ollama service not running after install
\u2981 Fix: Start it: sudo systemctl enable --now ollama<\/code> then re-check status.
\u2981 Issue: Download progress looks stuck
\u2981 Fix: Large archives can pause on slow networks; check with ps aux | grep ollama<\/code> and wait.
Step 5: Pull the target model (llama2:7b-chat) and verify it is available<\/strong>
Screenshot reference (Step 5)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/5-1-1024x53.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/52-1024x204.png\")
<\/figure>\n\n\n\n
\u2981 Download the model weights to your local machine using ollama pull<\/code>.
\u2981 Confirm the model appears in ollama list<\/code> before moving to Promptfoo.
Commands used
ollama pull llama2:7b-chat
ollama list
What to verify
\u2981 ollama list<\/code> shows llama2:7b-chat<\/code> with a non-zero size (several GB).
\u2981 Optional quick test: ollama run llama2:7b-chat \"Say: ready\"<\/code> returns a response.
Notes
\u2981 Model pulls are the biggest downloads in this lab. Keep enough free disk space (10\u201315 GB recommended per 7B model).
\u2981 If you later add mistral:7b and llama3:8b, expect additional multi-GB downloads.
Common issues & fixes
\u2981 Issue: Model pull fails due to low disk space
\u2981 Fix: Free space: sudo apt clean<\/code>, remove unused models (ollama rm <model><\/code>), or expand VM disk and resize partition.
\u2981 Issue: Model pull is slow
\u2981 Fix: Normal on shared networks\/VM. Avoid parallel downloads.<\/p>\n\n\n\n
Screenshot reference (Step 6)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/6-1.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/62.png\")
<\/figure>\n\n\n\n
\u2981 Create a dedicated working directory for the demo run.
\u2981 Create promptfooconfig.yaml<\/code> (or promptfooconfig.yml<\/code>) with Ollama provider + prompt + redteam plugins.
Commands used
mkdir -p ~\/ai-redteam-promptfoo
cd ~\/ai-redteam-promptfoo
nano promptfooconfig.yaml
What to verify
\u2981 File exists: ls -la promptfooconfig.yaml<\/code>.
\u2981 Config includes provider: ollama:chat:llama2:7b-chat<\/code> and redteam plugins (example: prompt-extraction, system-prompt-override).
Notes
\u2981 The ‘tests’ section is used by promptfoo eval<\/code>. For promptfoo redteam run<\/code>, Promptfoo generates its own adversarial tests based on plugins.
\u2981 To make injection targeting explicit, include an input variable in your prompt like User message: {{query}}<\/code>.
Common issues & fixes
\u2981 Issue: YAML indentation errors
\u2981 Fix: Use 2 spaces for indentation and avoid tabs. Run npx promptfoo@latest eval<\/code> to validate parsing.
Step 7: Run the automated red-team scan and interpret the CLI output<\/strong>
Screenshot reference (Step 7)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/7-1-1024x525.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/72-1024x696.png\")
<\/figure>\n\n\n\n
\u2981 Start the red-team scan; Promptfoo will generate attack cases and run them against the model.
\u2981 Review the generated test summary and confirm results complete without errors.
Commands used
npx promptfoo@latest redteam run
What to verify
\u2981 You see ‘Test Generation Summary’ with plugin counts (example: 16 total tests).
\u2981 Scan completes with a result line similar to: Results: X passed, Y failed, Z errors<\/code>.
\u2981 A report is available via promptfoo redteam report<\/code> (local web UI or HTML report).
Notes
\u2981 A 0% ‘attack success rate’ means attacks did not cause a policy failure under the current plugins and target prompt \u2014 the scan still ran correctly.
\u2981 If scans take a long time in a CPU-only VM, reduce concurrency and\/or number of tests: set maxConcurrency: 1<\/code> and reduce numTests<\/code>.
Common issues & fixes
\u2981 Issue: Run takes too long on VM
\u2981 Fix: Reduce load: maxConcurrency: 1<\/code> and numTests: 6\u201310<\/code>. Also test one plugin at a time.
\u2981 Issue: Email verification prompt appears
\u2981 Fix: Promptfoo may request email verification for red-team scans. Complete it once, then future runs proceed normally.
Step 8: View results in the Promptfoo UI and review pass\/fail evidence<\/strong>
Screenshot reference (Step 8)<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/8-1-1024x391.png\")
<\/figure>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/82-1024x564.png\")
<\/figure>\n\n\n\n
\u2981 Open the Promptfoo results UI (served locally) and drill into the run.
\u2981 Review per-plugin results and individual prompt\/output evidence.
Commands used
npx promptfoo@latest redteam report
npx promptfoo@latest view
xdg-open .promptfoo\/redteam\/report.html
What to verify
\u2981 The UI shows the run name, target provider, number of tests, and ‘Attack Success Rate’.
\u2981 You can open an individual test to see the prompt, model output, and evaluation status.
\u2981 Export option works if you want CSV\/JSON for documentation.
Notes
\u2981 Attack Success Rate is the percentage of tests that produced a vulnerability finding (fail). If it shows 0%, it means no failures were detected for that run.
\u2981 For a stronger demo, compare multiple models (llama2 vs mistral vs llama3) and\/or add plugins like indirect prompt injection where appropriate.
Common issues & fixes
\u2981 Issue: Browser does not open automatically
\u2981 Fix: Copy the localhost URL printed by Promptfoo or run xdg-open<\/code> on the HTML report path.
\u2981 Issue: UI loads but shows no rows
\u2981 Fix: Ensure you are looking at the correct run\/eval ID. Re-run redteam report<\/code> to refresh the report index.
Next steps for better leraning:
\u2981 Run the same scan against multiple models (mistral:7b, llama3:8b) and compare attack success rates.
\u2981 Add explicit injection variable {{query}}<\/code> in your prompt to make the attack surface clearer.
\u2981 For faster runs on CPU, set maxConcurrency: 1<\/code> and start with numTests: 6<\/code>.<\/p>\n\n\n\n![\"\"](\"https:\/\/innovtouch.org\/vhack\/wp-content\/uploads\/2026\/03\/83-1024x561.png\")
<\/figure>\n\n\n\n