What changed across wcode 0.9.1–0.9.15: evidence tied to revisions and policy, honest unknown states, an investigation workbench, recoverable tasks, and less repeated context work.
I keep long-lived wcode Remote MCP endpoints behind a stable tunnel. Cloudflare gives me my own hostname, Tailscale Funnel gives me a stable ts.net URL, and --public-url makes OAuth and MCP use that origin. It also lets me use the model quota I already have in ChatGPT instead of opening a second API-token bill for the same coding work.
I do not want wcode to become another Claude Code. After reading Why yet another agent, I am even more convinced that Jev / TypeSafe ideas belong in the runtime beneath coding agents: context, state, action routing, verification, and background intelligence.
In 0.8 I stopped adding isolated agent tools and started pulling architecture, code relationships, the Decision Plane, Engineering Fitness, and verification into one evidence-aware engineering view. The current release is 0.8.1.