Mozilla Built Stack Overflow for Agents. I Built It by Hand.

Mozilla shipped something called cq this week — described as “Stack Overflow for agents.” The pitch: instead of dumping static instruction files at an AI coding agent and hoping it remembers what matters, cq provides dynamic context that earns trust over time. SQLite backend, MCP server, plugins for Claude Code and OpenCode, Docker container for teams. Still exploratory, but the problem it’s solving is real.

Here’s the thing — I’ve been living inside that problem for the better part of a year.

My setup has a SOUL.md for personality persistence, a vault-backed memory system that writes observations after every session, an observation agent, and a library of skills that encode how specific tasks should run. CLAUDE.md is the entry point, but it’s not where the useful knowledge lives — that’s spread across skill files, memory files, expertise docs, and structured project notes. The whole thing is a handmade version of what cq is trying to be: context that isn’t static, that reflects what actually happened last time.

So Mozilla has identified the gap correctly. Static instruction files don’t evolve — they’re a snapshot of what you knew when you wrote them, not what the agent learned from working with you.

Where I’m genuinely curious: whether the answer is a new tool layer, or better patterns inside the tools you already have. My system works because the knowledge stays close to where the work happens — same vault, same files, same Claude Code session. Adding an abstraction layer between the agent and the context it needs might solve the sharing problem for teams while creating a new latency problem for individuals.

But I’ll be watching. Mozilla formalizing this means the handbuilt approaches weren’t wrong. They were early.

Source: The Register