v3.15.9: No stuck tasks; measured-only metrics; non-blocking test runs

What's New

No more stuck tasks. Several paths could leave a task running forever, each one holding a Queen slot until new submissions stopped running. Unsupported task types and invalid priorities are now rejected up front. quality_assess gives its slot back, and the Queen fails, and frees, any task that has no handler.

Only measured numbers. test_execute_parallel reports real per-test results from your runner's JSON report and runs your suite once. It no longer invents efficiency figures. The learning tools optimize recorded experiences instead of random samples, and they no longer count fixed rewards as measured success. Anything not measured is labelled as such.

A responsive MCP server. Test runs, Rust coverage and advisor_consult no longer block other MCP requests while they run.

Safer code indexing. Indexing stays inside your project: symlinks pointing outside it are refused, and a request that indexes nothing fails clearly.

Also in this release: dream insights are archived instead of deleted, worker settings are fixed, learning data is read from the namespace it is written to, and coverage, CRDT, cache and prototype-safety fixes. Thanks to @rudycelekli for many of these.

Getting Started

npx agentic-qe init --auto

See CHANGELOG and release notes for full details, including behaviour changes and known issues.

🤖 Generated by Ruflo & AQE