General AI coding tool compatibility layer

[darrenhinde]

Create a compatibility layer that allows OpenAgents to work with multiple AI coding tools (Cursor, Claude Code, Windsurf, etc.).

Vision

• Universal agent definitions
• Tool-agnostic context files
• Adapter pattern for different tools
• Unified workflow across tools

Challenges

• Different tool architectures
• Varying capability sets
• Maintenance overhead
• Testing complexity

Potential Approach

• Define common agent interface
• Create tool-specific adapters
• Shared context format
• Plugin architecture

[funnyfisherman]

"Hi @darrenhinde ,

Following up on my comment in issue #141, I wanted to drop a note here since this seems to be the main tracking issue for the VSCode extension.

I've been trying to integrate OAC into my workflow because I genuinely believe the 'architectural harnessing' and 'approval gates' approach is vastly superior to the scraping/soft-prompting methods used by other tools. However, as a non-coder/technical manager, the pure CLI environment is a significant cognitive blocker for me.

To try and bridge this gap, I actually spent some time building a 'Franken-OAC' workaround: a split-screen terminal wrapper (Kitty + VSCode) to force OAC into a more visual workflow. I've documented the whole experiment (and why it ultimately failed for me) in this repository:
👉 https://github.com/funnyfisherman/kitty_oac_addon

I'm sharing this to highlight just how much users like me (and potentially thousands of others who rely on VSCode/VSCodium) are eager for this native integration. A visual interface for the true architectural harnessing of OAC—specifically the MVI, the Context Scout, and the ability to enforce strict patterns rather than just relying on wishful prompting—would be an absolute game-changer.

I know this is in the backlog, but is there any rough ETA or timeline for when this might move forward? I'm not a coder, so I can't contribute directly to the extension's codebase, but I'm more than happy to help with testing, UX feedback, or documentation when the time comes!

Thanks again for the amazing work on the core framework."

[funnyfisherman]

One final thought on why this matters:

I know my request might seem a bit counter-intuitive: asking for a hardcore, developer-focused tool like OAC to be packaged into a highly user-friendly IDE extension.

It might seem counter-current because the general public is already being served by a flood of "zero-setup, zero-headache" tools like Anthropic's Claude Cowork or the viral OpenClaw. But these tools achieve their ease of use by demanding unbridled access to our entire workspaces, local filesystems, and personal apps.

Right now, we have "mom and pop" shops and non-technical founders letting their nephews install OpenClaw on their Macs with a credit card, giving an autonomous AI full access to their business data. We are handing out tools from the future with the security mindset of 1991. These "easy" tools are a massive danger to people's wallets, their computers, their businesses, and frankly, to humanity's control over its own systems.

But why do even smart developers fall for these dangerous tools? It comes down to cognitive exhaustion and the sociology of abandonment.

While the "Human-in-the-Loop" (HITL) model is universally recognized as the safest approach, it comes with a massive cognitive tax if the UX is poor. When a user has to spend 60% of their mental energy wrestling with an IDE and tracking AI deviations in their head, discouragement sets in rapidly. At that tipping point, 9 out of 10 users will abandon security for convenience. They will surrender to the "stupid but easy" autonomous tools simply because they are tired and just want to get things done.

Darren didn't invent the Human-in-the-Loop concept, but OAC is one of the very few frameworks applying a high-quality, strict architectural harnessing to enforce it, rather than relying on weak soft-prompting.

I am making a plea: for this quality harnessing to achieve mass adoption, it must become as user-friendly as the dangerous tools it ultimately protects us from. This is where the math becomes critical: The VSCode/VSCodium ecosystem represents over 15 million developers worldwide. Bringing a native, low-cognitive-load visual interface to OAC doesn't just help a few CLI-averse managers like me. It has the market potential to intercept millions of developers, founders, business executives, entrepreneurs, tinkerers, hobbyists, backyard hackers, and your grandma before they cross the tipping point and surrender their systems to unrestricted AIs.

This isn't just about making a better coding tool; it's about pushing a safe, sustainable paradigm into the mainstream before we all get swallowed raw by the OpenClaws of the world.