v3.10.4: Keep project learning in its own .agentic-qe + stop orphaned MCP busy-loop

What's New in v3.10.4

Two correctness fixes that keep your project's self-learning where it belongs and stop orphaned MCP servers from burning CPU — plus opt-in support for running your own local pattern hub and LLM judge.

🛠 Fixed

• Your project's learning now stays in its own .agentic-qe/ (#516). Previously, an .agentic-qe store in a parent folder (for example ~/.agentic-qe, which any aqe run from your home directory creates) would silently hijack every project beneath it, draining their learning into the wrong place. AQE now always resolves to the nearest project store, and the vector substrate (patterns.rvf/brain.rvf) no longer scatters stray stores when a command runs from a subfolder.

• Orphaned aqe-mcp no longer pins your CPU (#513). When a Claude Code session exited, its MCP server could be left running at 70–98% CPU indefinitely, ignoring normal termination. The server now shuts down cleanly the moment its parent goes away.

✨ Added (opt-in)

• Bring your own local pattern hub + LLM judge. Point the cross-project knowledge hub at a local nagual serve instance (NAGUAL_URL) and score pattern quality with a local Ollama-compatible model (NAGUAL_JUDGE_URL / NAGUAL_JUDGE_MODEL). Fully disabled unless you configure it — no change for existing setups.

Getting Started

npx agentic-qe init --auto

See the CHANGELOG and release notes for full details.