fix(dx): resolve packaging version and CLI advertising drift (MUT-035, MUT-050, MUT-051)

[BasilZafar11]

Fixes #78

Summary

• Make the repository root an unversioned uv workspace. Move contributor dependencies to the default dev group, retain optional LLM dependencies in llm, and remove the duplicate root CLI entrypoint. Regenerate the lockfile and update CI, contributor instructions, sample setup, and support templates to uv sync --group dev.
• Read Hosted API version from mutiny-api distribution metadata, with an explicit unreleased source fallback. Document the Hosted API as unreleased with an independent internal version, without changing release versions.
• Replace the stock web README with Mutiny Hosted setup, API proxy/auth configuration, Windows instructions, and checks.
• Remove the CLI Python 3.13 classifier so advertised minor versions match CI's 3.11/3.12 coverage. Clarify the source/workspace mutiny-api requirement in top-level and mutiny db --help output. CLI wheels still do not depend on or vendor the API.

Type of change

• Bug fix
• Tests
• CLI / DX
• Docs / community / DX

Checklist

• Followed CONTRIBUTING.md, security policy, support guidance, and code of conduct
• Linked related issue
• Core stays free of framework SDK imports
• Added regression coverage and wheel metadata assertions
• Documentation matches the implemented workspace commands
• One concern: packaging and contributor DX drift
• Authorized local/offline checks only

Test plan

• uv lock and uv sync --locked --group dev: successful, with no skipped root entrypoint warning.
• uv run --locked ruff check: passed.
• MUTINY_SAMPLE_OFFLINE=1 uv run --locked pytest -q --tb=short: 676 passed on local Python 3.13. CI continues to cover Python 3.11/3.12.
• Initial packaging/version regressions: 17 passed. Final packaging rerun with wheel metadata assertions: 13 passed.
• Sample offline mutiny init and mutiny run: successful; a violation regression was saved.
• mutiny --version reports 0.2.0; mutiny db --help states the API/workspace requirement. Unit tests also exercise help with API imports unavailable.
• Web typecheck/build and shell-based isolated artifact installation gate: left to CI; frontend changes are documentation only. Wheel builds and package integrity tests passed locally.

The setup migration from --extra dev to --group dev is documented in the changelog. Optional LLM setup uses --group llm. Branch starts from current upstream main.

[vercel]

@BasilZafar11 is attempting to deploy a commit to the priyanshu's projects Team on Vercel.

A member of the Team first needs to authorize it.

[BasilZafar11]

Opened PR #122 to address this issue. It removes the root product version, reads API version from metadata, documents Hosted development, removes the unsupported Python 3.13 classifier, and clarifies that mutiny db requires the source/workspace API package.

All 676 tests pass, along with Ruff, wheel checks, and offline sample smoke.

#122

[CodewithJha]

Thanks @BasilZafar11 — this is a careful, well-scoped packaging cleanup, and it lands every acceptance criterion in #78. Approving and merging.

What I verified on the branch

• Full suite: MUTINY_SAMPLE_OFFLINE=1 uv run --locked pytest -q → 676 passed; uv run --locked ruff check clean; mypy clean on mutiny_api/__init__.py and db_cmd.py.
• Offline sample mutiny run saves a violation regression; mutiny --version → 0.2.0; mutiny --help and mutiny db --help both state the source/workspace mutiny-api requirement.
• Bare uv sync (what Railway and docker-compose run) still installs every workspace member, so python -m mutiny_api keeps booting; the root no longer emits the "not packaged / scripts skipped" warning.
• Mutation check: reverting each fix to main individually (API __init__, CLI classifier, db help, root pyproject, web README) makes test_packaging_dx.py fail every time.
• The new web README matches the code: next.config.ts default MUTINY_API_URL, the middleware's server-side bearer injection, and serve.py's fail-closed non-loopback binds.

What I liked

• Making the root a virtual workspace with dependency groups fixes the version drift at its source instead of just bumping 0.1.0 → 0.2.0. It also removes the duplicate mutiny entrypoint and the redundant openai-agents extra.
• The wheel-metadata assertions in test_package_release.py (no 3.13 classifier, no Requires-Dist: mutiny-api, no vendored mutiny_api/) check the shipped artifact itself, not just the pyproject.
• Treating the API as "unreleased, independent version" is a valid answer to the first criterion, and it's documented in both READMEs.
• Honest test plan: it says what you left to CI.

Non-blocking follow-ups (no action needed on this PR)

• Because the workspace members now live in the dev group, bare uv sync on Railway/compose also installs pytest, pytest-asyncio and ruff into the Hosted runtime. That's harmless, but if #63 (compose/Railway hardening) is picked up, splitting runtime members into their own group and running uv sync --locked --no-dev --group <runtime> there would keep the runtime lean.
• uv.lock moved from revision = 3 to revision = 5 (newer uv). CI and Dependabot handle it. Contributors on an older uv may need to upgrade before --locked passes.

[CodewithJha]

Merged in a271366, thanks again @BasilZafar11. #78 is now closed and indexed as shipped. If you'd like a next one that fits the packaging, CLI and test work you did here:

• [TEST] MUT-027, MUT-048, MUT-052: CLI, reliability, and adapter extract tests miss product paths #75 (MUT-027, MUT-048, MUT-052): CLI/loader unit coverage and product-path tests. You already wrote solid CLI help and wheel-metadata tests here.
• [DOCS] MUT-030, MUT-031, MUT-032, MUT-033, MUT-064: Docs lag 0.2.0 release and observe-only Hosted #77 (MUT-030–033, MUT-064): docs lag the 0.2.0 release. A natural follow-on to the README and CLI.md cleanup in this PR.
• ci: run mutiny test in sample project smoke (seed or discover regressions) #28: run mutiny test in the sample-project CI smoke. A small CI change that builds on the offline sample flow you exercised.

No pressure, and comment on the issue first so nobody duplicates the work.