RedWidowPentest in VS Code, with AI on call. Kali · Exegol · Terminals · Kill chain · Optional AI agent A workspace for your CTFs, labs and authorized pentests. Prepare containers, open terminals and bring your findings together. Work by hand or call an agent when you need one. Previously Venator Ops / Warband. The update keeps your projects, conversation history and settings. Your workspace in one place
The cockpit, terminals, manual evidence capture and exports work without an AI connection. Get startedYou need VS Code 1.90+ and Docker installed and running for containers. VSIX packages are available for Windows x64 and Linux x64. You do not need Node.js or a development environment.
Providers include Claude, OpenAI / ChatGPT and OpenCode with its full runtime provider catalog. Search and enable providers such as Google Gemini, OpenRouter, GitHub Copilot and DeepSeek in Providers. The general OpenCode entry shows all connected models. Use Add Ollama / a compatible endpoint to enter your API URL and installed model IDs. You can switch provider while keeping the same conversation. Provider subscriptions and API charges are separate. On Windows, Claude requires Git for Windows with Git Bash. You choose a shared folder when you create a container. A container is linked to the active project only if the project folder is inside that shared folder; otherwise it is created unlinked, which is not an error. An active project is not required to create a container; without one, the container is not linked. Files in the shared folder appear directly in the VS Code Explorer. Preparing a linked container also provides a named path under In Settings → Appearance → Docker terminal prompt, choose Original, RedWidow, Compact or Minimal, with a preview of each. The choice applies to new Bash terminals opened by RedWidow in Kali or Exegol. Existing terminals keep their prompt, and no English is the default interface language. Choose EN / FR in Settings. Native command titles follow your VS Code language. Add context and toolsThe chat’s + menu brings together files, container mentions, terminal context, commands, Monitor and kill chain shortcuts. Use
Attach text, source files, logs and PNG, JPEG, GIF or WebP images with the file picker, drag and drop, or paste from the clipboard. Preview images and remove attachments before sending. Up to 10 attachments: text files are limited to 64 KiB each / 256 KiB total, images to 5 MiB each / 12 MiB total. The composer warns when image support is unknown and blocks image sends when the selected model explicitly lacks vision. PDFs and other binary documents are not supported. Clipboard file access depends on the platform; pasting a path does not read that file. In Settings → Project → Integrations, enable the method profiles and MCP connectors you want available. Configure each external MCP service for your project. Then use + to choose one method, up to four MCP connectors, Burp and web search for this conversation. Disabling a connector preserves its configuration. A new conversation starts without optional resources selected. Methods include Mentor, Bug bounty, Out of the box, Red team and HTB / CTF. A short brief guides the agent; detailed methods load only when needed. External MCP programs must be installed and started separately. Learn with hintsPut Markdown or PDF course material in the folder opened by Settings → Atlas → Open courses folder. Enable Learning in the chat’s Configure menu, attach terminal context with Stay in controlSensitive actions produce approval cards. Allow an action once, deny it, or remember an eligible approval for the current conversation. Passwords, tokens and keys are masked before they reach logs. Bypass approves actions for a time you choose, from 1 minute to 48 hours, or without limit, including actions the gate would normally deny. Its active state remains visible. Turn it off with the chat switch, the command RedWidow: Turn off the approval Bypass, or by clicking the status bar. It also switches off when the engine stops. Use it only for a scope you control. Learning mode is excluded. Red team engagementRun RedWidow: Open the engagement to import and validate the engagement sheet and the campaign plan (Validate engagement…, Validate plan…). Nothing runs without a human validation of their fingerprint. Emergency stop › Activate… makes the agent's network and shell actions against the scope be refused; Lift the stop… releases it. The stop is saved on disk and survives closing VS Code. HistoryHistory is shared by all projects. It is read-only; resuming a conversation from another project asks you to switch to that project first. History left under the former UninstallRun RedWidow: Clear token, delete your containers in the Cockpit, then uninstall the extension. Some data stays on disk: MoreUser guide and source · Issues · MIT license |