v2026.7.13

Added

• Added a bundled persistent-kernel eval extension for JavaScript, Python, Ruby, and Julia, with persistent state, agent() and output() bridges, structured status streaming, bounded spillable output, rich terminal rendering, completion schemas, and cancellation-safe cleanup.

Changed

• Reworked the GPT-5.6 prompt preset into a Hephaestus-parity autonomous deep worker with explicit manual QA, failure recovery, shared-worktree safeguards, and outcome-based stop rules.

Fixed

• Fixed post-compaction queued input and tool continuations so they no longer deadlock, lose ownership, or surface stale continuation failures while the TUI and neo resume work.
• Fixed project-trust reloads dropping builtin and bundled extensions, including the bundled eval extension.

Removed