*This is a submission for the GitHub Copilot CLI Challenge*
What I Built
I built the Agent OS VS Code Extension, a dedicated development environment designed to make building AI agents safe, compliant, and observable by default.
As we move from chat-based LLMs to autonomous agents, we need better tooling to manage the "governance layer"—the policies, permissions, and safety checks that prevent agents from going rogue. This extension brings those controls directly into the IDE.
Key Features:
- Visual Workflow Designer: A drag-and-drop interface to architect agent flows (exportable to Python/TypeScript) without getting lost in spaghetti code.
-
Real-time Security Diagnostics: Think of it as a linter for agent safety. It flags dangerous patterns (like unconstrained
os.systemorevalcalls) and suggests safer alternatives as you type. -
Policy Editor: A GUI for defining strict boundary conditions (e.g., "Agents cannot write files outside
/tmp" or "No external API calls to non-allowlisted domains"). - Metrics Dashboard: Embedded telemetry to watch agent activity and policy violations in real-time.
It’s designed to bridge the gap between "prototype" and "production-safe" for AI engineers.
Demo
You can find the full interactive documentation, including the extension download and tutorials, here:
Agent OS VS Code Extension
(Note: The page includes interactive previews of the Policy Editor, Workflow Designer, and Security Diagnostics.)
My Experience with GitHub Copilot CLI
Building a VS Code extension requires juggling a lot of disparate tools: the vsce packaging tool, complex Git flows for release management, and the VS Code API itself. I used the GitHub Copilot CLI to speed up the "glue" work that usually breaks my flow.
1. Untangling the VS Code API
The VS Code extension API is massive. Instead of context-switching to browser documentation when I forgot how to register a specific Webview provider, I used the CLI to get quick scaffold explanations directly in my terminal. It saved me from the "documentation tab fatigue" typical of extension dev.
2. Streamlining Release Management
Publishing an extension involves strict versioning and packaging steps. I used gh copilot suggest to generate the correct shell commands for tagging releases and running vsce package with specific flags, ensuring I didn't mess up the build artifacts.
3. Regex for Security Rules
One of the core features of Agent OS is the "Security Diagnostics" linter. Writing the regex to robustly catch things like subprocess.call(..., shell=True) without false positives is tricky. I used the CLI to brainstorm and refine these regex patterns, which significantly accelerated the implementation of the diagnostics engine.
Overall, the CLI acted as an intelligent second monitor that lived right in my terminal, keeping me focused on the architecture rather than the syntax of the tools I was using.
Top comments (1)
The extension is built on top of the Agent OS kernel: github.com/imran-siddique/agent-os