Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Addresses code-review feedback: CliRunner/app are already imported at
module scope in each test file; remove the duplicated in-function imports
and normalize blank lines.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Addresses code-review feedback: replace constructor-only contract tests
with assertions on the real raise sites (flow.run_login_flow timeout,
client.resolve_audio_source), checking message and suggestion separately.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Golden tests pin the exact rendered CLI surface: every leaf command's --help
(usage + options + Examples epilog) and the human/JSON forms of errors
(including the Suggestion: line). The command list is derived from the live
Typer tree, so new commands are automatically required to have a snapshot.
Refresh intentionally with: pytest --snapshot-update

Requires the syrupy>=4.0 dev dependency.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Parametrized over every init template: build a clean venv with uv, install
its requirements.txt, and import api/index.py in that venv. Catches the one
drift the static contract tests can't — a dependency listed but uninstallable,
a missing transitive pin, or an import that only fails with the real packages
present.

Gated behind a new `install` marker (slow; needs network + uv) and excluded
from the default run and precommit gate; skips rather than fails when uv is
absent or PyPI is unreachable.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add pyright as a dev dependency and gate on it in scripts/check.sh
(after mypy), scoped via a new [tool.pyright] section to the same
aai_cli + tests sources mypy already covers. pyright resolves
third-party types mypy ignores, so it surfaces a stricter class of
issues; fix all 120 it reported so the gate lands green.

Source fixes:
- config_builder: type the dynamic SDK-kwargs dicts as dict[str, Any]
  at the construct_* unpack sites (the honest type for values bound
  for the SDK constructor).
- microphone: broaden stream_factory to Callable[..., Iterable[bytes]]
  — the source only iterates the result + getattr(close), never needs
  Iterator.
- scaffold: import Traversable from importlib.abc to match the protocol
  resources.files() actually returns.
- transcribe template: guard the None transcript id before returning.
- youtube: scoped ignore for yt-dlp's private _Params TypedDict.
- llm/auth: tighten Any returns to ChatCompletion / Mapping now that
  pyright sees the real types.

Test fixes: annotate TypedDict literals, narrow Optional fakes via
local + assert, type stand-in modules as Any, and cast duck-typed
stand-ins (CLIError, ChatCompletion, TextIO, Generator) to the API
type they impersonate. Add numpy dev dep (e2e resampling) and wrap
the over-long stream example line.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Consolidate the two duplicate Typer-tree walkers into a shared
tests/_cli_tree.py and drop the ~12 per-command *_help_has_examples
tests now subsumed by the tree-derived examples-coverage guard plus the
full --help snapshot. Remove redundant inline imports / CliRunner()
re-instantiation in the error-suggestion tests, and cache the install
test's PyPI reachability probe.

Also add coverage for AMS auth, init runner, config_builder error
paths, and duplex audio, and tidy output.emit_error's suggestion render.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- Reformat output.py (collapse the Suggestion: print onto one line).
- Exclude tests/__snapshots__/ from trailing-whitespace/end-of-file-fixer:
  syrupy stores .ambr blank lines as "  ", so trimming them desyncs the
  goldens from the live render and breaks all snapshot tests in CI.
- Bump ruff-pre-commit to v0.15.16 to match the project's ruff dev pin.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

`uv run` does not install [project.optional-dependencies] by default, so the
uv-managed env lacked pytest/hypothesis/fastapi/numpy/etc. mypy hid this via
ignore_missing_imports, but pyright reported 44 reportMissingImports errors and
failed CI. Run the type-checkers and pytest with `uv run --extra dev`.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

GitHub runners set FORCE_COLOR, which overrides NO_COLOR in Rich, so the
--help renders carried SGR escape codes the plain goldens don't have — 17
snapshot tests failed only on CI. _normalize now strips ANSI before comparing
(no-op for the already-plain error renders), making the goldens color-env
independent.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Move dev deps from [project.optional-dependencies].dev to
[dependency-groups].dev, which uv installs by default for uv run/uv sync.
This drops the `uv run --extra dev` band-aid in check.sh and makes a plain
.venv correct for editors, so [tool.pyright] is pinned to .venv — Pylance and
the CLI now resolve imports identically.

The code already required 3.11 (config.py imports stdlib `tomllib`, added in
3.11), so requires-python, the mypy/pyright/ruff targets, doctor's runtime
gate, install.sh, and the README all move to 3.11 — making pyright resolve
tomllib under the new venv pin. CI installs the group via
`pip install --group dev` (needs pip >= 25.1; upgraded in the pre-commit job);
the check.sh job just needs uv on PATH.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>