Bandit Stealth connects VS Code directly to the Bandit API so engineers can send prompts, launch autonomous plans, and review diffs without leaving the editor.
Key Features
Streaming chat with prompt history, intent chips, and quick follow-ups.
Agent run dashboard with live step status, telemetry, and goal evaluation.
Diff review loop that stages edits, shows inline summaries, and supports dry-run mode.
Built-in API key management plus mode/model selectors for Ask vs. Agent workflows.
Requirements
VS Code 1.75 or later
Active Bandit API key (bai_…)
Python 3.8+ on PATH for agent helpers
Getting Started
Install the extension from the VS Code Marketplace (or drag the .vsix into VS Code).
Open the Bandit Stealth view from the Activity Bar.
Run Bandit Stealth: Set API Key and paste your bai_… credential.
Start chatting or launch an agent goal—Stealth handles context capture, execution, and diff review from there.
The extension automatically checks for Python 3. If it cannot find one, it prompts you to install Python or set banditStealth.pythonPath.
Configuration Highlights
banditStealth.model: default model slug sent to the API.
banditStealth.apiUrl, banditStealth.intentUrl, banditStealth.feedbackUrl: service endpoints.
banditStealth.insertResponsesByDefault: toggle auto-inserting completions into the editor.
See VS Code settings (Cmd/Ctrl+, → search “Bandit Stealth”) for the full list.
Support & Contributions
Bandit Stealth is an internal Burtson Labs project. External contributions aren’t accepted right now, but feel free to reach out to team@burtson.ai if you have feedback or want to discuss integrations.
