Add Velaria Agent TUI

[ashione]

Summary

• Add Velaria-owned Agent CLI/TUI entry points with Textual, including default no-args Agent mode, -i compatibility, headless --print/--stream-json, and --model overrides.
• Add a responsive TUI shell with streaming Markdown rendering, running/queued state, history scrolling, keyboard shortcuts, and model switching for Codex/Claude runtimes.
• Remove the old prompt-toolkit REPL path and update docs, skills, dependencies, and tests around the new Agent-first CLI flow.
• Preserve existing finance CLI registration from main while rebasing the PR to TUI-only scope.

Scope

• This PR now contains a single commit: cfefb13 Add Velaria Agent TUI.
• The previous Float32 / SIMD / C++ core commit has been removed from this PR branch and preserved locally as auto/velaria-agent-tui-with-float32.

Validation

• uv run --project python python -m unittest python.tests.test_agent_cli python.tests.test_python_cli_contract python.tests.test_ai_runtime_agent
• bazel test //python:agent_cli_test //python:python_cli_contract_test //python:ai_runtime_agent_test
• uv run --project python python python/velaria_cli.py --help
• uv run --project python python python/velaria_cli.py agent --help
• uv run --project python python python/velaria_cli.py (non-TTY structured error path)
• git diff --check origin/main...HEAD
• legacy REPL/prompt-toolkit residual scan
• changed-file secret pattern scan (only placeholders, test keys, variable names, and lockfile package names)

Note

• python.tests.test_claude_runtime was not included in the final validation because it currently fails on this checkout with a pre-existing fake claude_agent_sdk module lacking __file__; this is outside the TUI-only diff.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b247f1b12f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Propagate failed agent turns from --print

When the Codex or Claude adapter reports a runtime/auth/tool failure, it yields an error or turn.failed event instead of raising; this loop only collects assistant_text, then run_agent_print always prints the accumulated text and returns 0. In scripted velaria_cli.py agent --print ... usage, a failed turn can therefore appear as a successful empty or partial answer, so callers cannot detect or retry the failure.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Make the missing-Textual fallback importable

If a user has not run the updated Python dependency sync yet, importing Textual raises ModuleNotFoundError, the fallback assigns ModalScreen = _MissingWidget, and this class definition immediately evaluates ModalScreen[str | None]. _MissingWidget is not subscriptable, so tui_app raises TypeError during import before run_tui can show its intended “requires Textual” message, causing the new default CLI entry to crash in exactly the unsynced environment this fallback is meant to handle.

Useful? React with 👍 / 👎.