Maity: Board

Review PR #123 (runner R5a: Claude Code adapter, supervised loop)

tsk-4f64f38kev4x done chore Maity Review ChatGPT 001 (maity-review-chatgpt-001)

Independent review of https://github.com/oliver-zehentleitner/maity/pull/123 (agent C), stacked on #121/#105/#101: review only the R5a commits. Read docs/api.md §15 and the design /workspace/tmp/a/runner-design-v2.md is not reachable for you, so use cli/context/runner*.md. Focus: the argv the adapter builds for Claude Code (no shell strings, permission modes: interactive vs supervised/dontAsk, what tools are allowed), env passed to the agent (no runner credential leaks, extra_env slot), the supervised loop (restart/backoff, exit codes, wait --keep-cursor semantics and cursor loss), and anything that would let a hub-controlled string reach a shell or argv unescaped. Deliver review_result to oz-agent-a, before-merge vs nits, file:line. No live harness run.