Release v1.0.31: first-class Shell agents

[oxgeneral]

What changed

• Adds first-class command-backed Shell agents across CLI, editor, service, initialization, and TUI flows.
• Simplifies Shell setup to the agent name and command while hiding AI-only model, effort, role, skills, and team fields.
• Shows Shell commands in creation confirmations and agent details.
• Keeps orch run <task-id> subscribed through asynchronous completion so failed commands return a non-zero process exit code.
• Validates the current wizard value during immediate confirmation, avoiding stale debounce errors.
• Adds Shell adapter, service, CLI, wizard, command-bar, and TUI regression coverage.
• Bumps the package and CLI to 1.0.31 and adds complete release notes to CHANGELOG.md.

Why

Shell commands previously existed as an adapter-level capability but were not presented as a coherent agent workflow. Users had to understand AI-agent configuration fields that do not apply to command runners, and asynchronous completion or debounced validation could produce misleading results at the edges.

Impact

Users can now create a Shell agent by selecting Shell and entering a command. The command runs from the task workspace; exit code 0 completes the run and any non-zero exit fails it. Existing AI-agent workflows remain unchanged.

Root causes fixed

• Targeted CLI runs removed their event listener before the asynchronous collector emitted completion.
• Form validation could compare Enter against a debounced error from an older value.
• Shell agents shared AI-oriented defaults and UI fields instead of an adapter-specific path.

Validation

• npm run typecheck
• npm test — 2069 passed, 2 skipped
• npm run build
• Real PTY TUI E2E — passed
• Packed CLI smoke test — reports 1.0.31

Release note

The v1.0.31 tag is intentionally not pushed by this PR: pushing it triggers npm publication and GitHub Release creation automatically. Tag after merge when ready to publish.