fix(cli): address non-interactive and endpoint list failures

[cursor]

Summary

• avoid blocking on fine-tuning launch confirmation in non-interactive mode while preserving price and account-balance warnings
• restore tg beta endpoints list as the canonical command and retain ls as its alias
• add regression coverage for both telemetry-observed failures

Context

The daily CLI failure export contained an EOF when reading a line from fine-tuning create in a non-interactive process and an unknown-command failure for endpoints list. Both still reproduced on current main, and no existing PR covered them. The preview-rendering and LoRA-download failures in the same export were already covered by PRs #466 and #469.

Tests

• TOGETHER_TELEMETRY_DISABLED=1 .venv/bin/python -m pytest -n 0 tests/cli/test_fine_tuning.py tests/cli/test_beta_endpoints.py (27 passed)
• .venv/bin/ruff check src/together/lib/cli/__init__.py src/together/lib/cli/api/fine_tuning/create.py tests/cli/test_fine_tuning.py tests/cli/test_beta_endpoints.py
• .venv/bin/ruff format --check src/together/lib/cli/__init__.py src/together/lib/cli/api/fine_tuning/create.py tests/cli/test_fine_tuning.py tests/cli/test_beta_endpoints.py
• .venv/bin/pyright --pythonpath .venv/bin/python src/together/lib/cli/__init__.py
• .venv/bin/pyright --pythonpath .venv/bin/python src/together/lib/cli/api/fine_tuning/create.py

  

[broly-code-security-scanner]

Broly Security Scan

Note

✅ Clean scan
No vulnerabilities detected in this PR.

Note

Re-scan this PR anytime with /broly scan — useful after /broly undismiss, or to refresh findings without a new push.

Broly — SAST (zai-org/GLM-5.2) · Secrets · SCA · GH Actions (zizmor) · Containers · SBOM · Powered by Together AI

[blainekasten]

@cursoragent should we consider changing this to use the confirm method from src/together/lib/cli/utils/_prompt.py

Would that work?

[cursor]

Yes. It works, but alias it because the confirm CLI parameter shadows the helper:

from together.lib.cli.utils._prompt import confirm as prompt_confirm

if not confirm:
    console.print(get_confirmation_message(price_line=price_line, warning=warning))
    if not config.non_interactive and not await prompt_confirm("Do you want to proceed?"):
        return

This preserves blank-input-as-yes, avoids prompting in non-interactive mode, and handles EOF/cancellation as “no.” Keep the non_interactive guard; the helper alone would otherwise abort instead of submitting.

  

[blainekasten]

@cursoragent let's do that

[cursor]

Done. Fine-tuning now uses the shared async confirm helper while retaining the non-interactive guard.

Checks: 19 tests passed; Ruff and Pyright clean.

  

[blainekasten]

@cursoragent Can we have the command name stay as ls and have a hidden version that aliases list to this command? I'm ok with it working, but don't want the help output text to show the name list

[blainekasten]

@cursoragent Can we have the command name stay as ls and have a hidden version that aliases list to this command? I'm ok with it working, but don't want the help output text to show the name list

[cursor]

Implemented:

• ls remains the visible canonical command.
• list is registered separately with show=False.
• Added help-output regression coverage.
• Both commands still work.

Checks: 9 tests passed; Ruff and Pyright clean.

Environment lacked dependencies; consider an environment setup agent at cursor.com/onboard: “Install uv 0.10.2 and run uv sync --all-extras.”