… videos are never committed

tests/screencast/ was a raw copy of the demo-video generator from the
strictdoc-project.github.io (Hugo) repository. This is the first step of
moving that generator into the StrictDoc repo itself (see
developer/tasks/20260704_screencast/task.md).

Removed as part of this cleanup:
- demo.js, package.json, package-lock.json, node_modules/: the JS/Playwright
  pipeline is being rewritten in Python (Playwright has a full Python API),
  so StrictDoc gains no new Node.js/npm dependency.
- run_demo.py: only orchestrated the now-removed `npm run demo` step.
- fixtures/strictdoc-demo-project/.local/: personal working notes with
  another machine's absolute paths and stray duplicate .sdoc files, not
  part of the actual demo fixture.
- __pycache__/, .DS_Store, and the generated fixtures/.../output/ cache:
  build artifacts, not source.

Still present, to be replaced in a later stage: run_server.py and
strictdoc_server.py, which currently start StrictDoc by shelling out to
`tox` against a sibling repo checkout — an assumption that only made sense
outside the StrictDoc repo. These will be replaced by reusing
tests/end2end/server.py::SDocTestServer.

No functional change to StrictDoc itself; tests/screencast is not yet wired
into any invoke task.

Adds the `playwright` Python package to requirements.check.txt, alongside
seleniumbase, since it is installed and used through the same tox `check`
environment as the end2end test suite.

This is the dependency needed for the tests/screencast video-generation
pipeline being rewritten in Python (see
developer/tasks/20260704_screencast/task.md). The pipeline code itself and
the one-time `playwright install chromium` browser-binary setup step will
follow in later stages of that task.

Replaces the ad-hoc strictdoc_server.py (which shelled out to `tox -e
py310-development` against a sibling `../strictdoc` checkout — an
assumption that only made sense when this generator lived in the
strictdoc-project.github.io repo) with the already-hardened
tests/end2end/server.py::SDocTestServer, used as-is via its context-manager
interface.

Adds tests/screencast/fixture.py with the shared fixture-project and port
constants (a separate RECORD_SERVER_PORT is reserved for the upcoming
video-recording mode, so a manual dev server can stay open while a scenario
re-records).

run_server.py is rewritten to start/stop the demo server through
SDocTestServer directly — no tox subprocess, no sibling-checkout
assumption, no STRICTDOC_ROOT/STRICTDOC_TOX/pyenv-version env vars.

Manually verified: the server starts against the demo fixture, serves
HTTP 200, and shuts down cleanly on exit.

…st tests

Replaces the JSON-step scene definitions in the removed demo.js with two
ordinary pytest test cases, following the tests/end2end/screens/*/test_case.py
convention: each screencast scenario is now a real regression check, not
just a replayed sequence of UI actions.

- strictdoc_ui/test_case.py: browse a StrictDoc document, switch to the
  table view, and assert on document title, viewtype menu, and the
  resulting table view state.
- ide_typing_to_table/test_case.py: the IDE-style fake-typing scene
  (demo.html) followed by the same table-view flow.
- scenarios/conftest.py: a local `page` pytest fixture (headless Chromium
  via playwright.sync_api directly — no pytest-playwright plugin needed for
  this scope).
- scenarios/typing.py: the fake-typing effect ported from demo.js's
  typeText() (appends to textContent; not real keyboard input).

Note: tests/end2end's Screen_*/component Page Object helpers are tied to
SeleniumBase (BaseCase, By.XPATH) and cannot be reused as-is by Playwright
code. The same underlying selectors (#viewtype_handler,
[data-viewtype_link="table"], body[data-viewtype], .header__document_title)
are used directly instead, verified against the actually rendered demo
fixture pages.

Both scenarios currently run without recording video; recording mode is
added in a follow-up stage.

Manually verified: both tests pass under the `check` tox environment
(playwright + chromium installed) against the demo fixture server.

…st tests

Replaces the JSON-step scene definitions in the removed demo.js with two
ordinary pytest test cases, following the tests/end2end/screens/*/test_case.py
convention: each screencast scenario is now a real regression check, not
just a replayed sequence of UI actions.

- strictdoc_ui/test_case.py: browse a StrictDoc document, switch to the
  table view, and assert on document title, viewtype menu, and the
  resulting table view state.
- ide_typing_to_table/test_case.py: the IDE-style fake-typing scene
  (demo.html) followed by the same table-view flow.
- scenarios/conftest.py: a local `page` pytest fixture (headless Chromium
  via playwright.sync_api directly — no pytest-playwright plugin needed for
  this scope).
- scenarios/typing.py: the fake-typing effect ported from demo.js's
  typeText() (appends to textContent; not real keyboard input).

tests/end2end's Screen_*/component Page Object helpers are tied to
SeleniumBase (BaseCase, By.XPATH) and cannot be called from Playwright
code. Instead of inlining locators per test, this adds new Playwright-native
helpers under tests/screencast/helpers/ (Screen, ViewTypeSelector),
mirroring the existing helpers' names, selectors, and responsibilities —
meant for reuse across this suite now and as a starting point for a wider
Playwright migration later, per SDG guidance to avoid one-off duplication
once more than one call site exists.

Both scenarios currently run without recording video; recording mode is
added in a follow-up stage.

Manually verified: both tests pass under the `check` tox environment
(playwright + chromium installed) against the demo fixture server.

Makes each screencast scenario dual-purpose, as designed in
developer/tasks/20260704_screencast/task.md: by default it's a fast
pass/fail regression check with no video output; with
--strictdoc-record-video, the same test also captures and saves a .webm.

- fixture.py: add OUTPUT_DIR (tests/screencast/output/).
- scenarios/conftest.py: add the --strictdoc-record-video pytest option;
  the `page` fixture enables Playwright's record_video_dir only when the
  flag is set, and on context close saves the recording as
  output/<scenario-dir-name>.webm (e.g. strictdoc_ui.webm), removing
  Playwright's temporary video file afterwards.

Removed the two leftover .webm files from the old JS pipeline
(strictdoc-ui.webm, ide-to-table.webm) — different naming scheme, no
longer produced by this generator.

Manually verified: scenarios pass both without the flag (no video written)
and with it (both named .webm files produced, no stray temp files left in
output/).

Adds `invoke test-screencast` (alias `tsc`) in tasks.py, following the same
tox `check`-environment convention as test_end2end/test_unit_server
(junit-xml under build/test_reports/, dedicated pytest cache dir).

Options:
- --focus: pytest -k filter, same as other test-* tasks.
- --record-video: forwards --strictdoc-record-video to pytest, so the same
  invocation both runs the regression checks and (re)generates the .webm
  videos into tests/screencast/output/.

Kept separate from invoke test-end2end per
developer/tasks/20260704_screencast/task.md: screencast runs are
slower/produce binary artifacts, and the scenarios always run headless (no
headed/--headless flag needed, unlike SeleniumBase-based end2end tests).

Manually verified: `invoke test-screencast` and
`invoke test-screencast --record-video` both pass and behave as expected
(2 passed; the latter produces strictdoc_ui.webm and
ide_typing_to_table.webm in tests/screencast/output/).

…line

Replaces the README carried over from strictdoc-project.github.io (which
documented `npm install --prefix demo`, STRICTDOC_ROOT/STRICTDOC_TOX env
vars, and a sibling-checkout setup) with docs matching the pipeline as it
now exists in this repo: directory layout, the one-time
`playwright install chromium` setup step, `invoke test-screencast`
(with --focus/--record-video), the manual dev server via run_server.py,
and a short guide for adding new scenarios using the helpers/ Page Objects.

A failing scenario means its video can no longer be trusted, not that it
should be re-recorded automatically: investigate first (it may be a product
bug or an intentional UI change the scenario doesn't reflect yet), and
re-record once the scenario is fixed or updated and passing again.

Also fixes fixtures/strictdoc-demo-project/README.md, which still referred
to a `demo/` directory that no longer exists.

Manually verified: `invoke test-screencast` passes (2 passed) as a final
end-to-end sanity check of the whole rewritten pipeline.

…plate

- Rewrites WHAT/WHY/HOW to match developer/tasks/task.template.md
- Removes narrative that didn't belong in a task specification

… dev port

…ctions

run_server.py imports project dependencies (psutil, watchdog via
strictdoc itself) that only exist in this project's tox environments, not
in a bare system `python3`. Running it directly with `python3
tests/screencast/run_server.py` fails with `ModuleNotFoundError: No module
named 'watchdog'`.

Adds `invoke screencast-server` (alias `scs`), following the same pattern
as the existing `invoke server` task: runs run_server.py through the `check`
tox environment (the one with psutil/playwright/etc., matching
`invoke test-screencast`), so it has the dependencies it needs.

Updates tests/screencast/README.md to document `invoke screencast-server`
instead of the bare `python3` invocation.

Manually verified: `invoke screencast-server` starts the demo server and
serves HTTP 200 against the fixture project; `invoke test-screencast`
still passes end-to-end after these changes.

…ults

DEV_SERVER_PORT was 5111 — StrictDoc's own default project server port
(ProjectConfigDefault.DEFAULT_SERVER_PORT), the same one `invoke server`
uses. Running `invoke screencast-server` and `invoke server` at the same
time (or leaving one running) collides on that port; this is what caused
`invoke server`'s "Address already in use" failure.

RECORD_SERVER_PORT (5112) had the same problem one level down: it's
tests/end2end's SDocTestServer default fallback port, used by any end2end
test that doesn't pass an explicit port.

Moves both ports to 5301/5302, clear of StrictDoc's own default (5111) and
the end2end port range (5112-5212). Updates tests/screencast/README.md's
port references to match.

Manually verified: `invoke test-screencast` (2 passed, server on 5302) and
`invoke screencast-server` (serves HTTP 200 on 5301) both work on the new
ports, and leave 5111 untouched.

…-screencast

Adds .github/workflows/screencast-tests.yml, separate from
end2end-tests.yml: sets up the `check` tox environment, installs
Playwright's Chromium binary (not covered by pip install), then runs
`invoke test-screencast`. Without this, the WHY in
developer/tasks/20260704_screencast/task.md ("ties video freshness
directly to the same CI/dev signal already used for UI regressions") was
aspirational rather than true — nothing ran these scenarios automatically.

Also fixes lint issues in tests/screencast/run_server.py surfaced by
`invoke lint`: dropped the unused shebang (the script is only ever invoked
via `python tests/screencast/run_server.py`, never executed directly),
switched `List[...]` to `list[...]`, and added `# noqa: T201` to its
prints, matching the rest of the codebase's convention for intentional
console output.

Manually verified: each workflow step run locally in sequence (tox
--notest, pip_install_strictdoc_deps.py, playwright install chromium,
invoke test-screencast) succeeds; `ruff check tests/screencast/` is clean.

- `invoke screencast-server` gains `--focus=<scenario>` to manually
  preview scenarios that don't use the shared fixture project (like
  hello_world, which generates its own project on demand into the
  gitignored build/screencast_manual/<scenario>/ and reuses it across
  restarts).

- New `hello_world` screencast scenario: runs a real `strictdoc new`,
  types the command and its actual stdout into a terminal-style HTML
  scene, then drives a live server through creating a document and a
  requirement via the web UI (tests/screencast/scenarios/hello_world/).
- New Playwright helpers ported from the e2e SeleniumBase suite for the
  add-document and add-requirement flows (tests/screencast/helpers/).

Adds Pointer (fake mouse cursor + click-target highlight, since
Playwright renders no real cursor) and pause() pacing helper; reworks
the hello_world scenario to skip document creation, add a Requirement
to the existing High-Level Requirements doc instead, and close with an
editor-style scene revealing the real appended .sdoc lines.

Moves the cursor/highlight CSS out of the inline JS template string in
pointer.py into pointer.css, embedded via json.dumps() instead of
repr() so the CSS can safely contain quotes/backslashes without
breaking the generated JS.

…DocLexer

Reuses StrictDoc's own Pygments lexer and pygments.css (same ones used
to highlight `.. code:: strictdoc` blocks in the docs) to render the
.sdoc source in the editor scene, instead of plain unhighlighted text.

…syntax highlighting in README

Adds coverage in tests/screencast/README.md for the helpers table, video resolution (VIEWPORT_SIZE), pacing/pause knobs, the fake cursor/highlight mechanism (pointer.py/pointer.css), editing terminal/editor scene styles, and the real StrictDocLexer-based syntax highlighting — none of which was documented after the recent Pointer/hello_world work.

…dd --edit and a startup banner

`invoke screencast-server` now defaults to a fresh, disposable copy of
the scenario's project (nothing persists, the shared fixture is never
touched); pass --edit to serve the real, persistent files instead.
Replaces the plain startup print with a boxed banner (matching
StrictDoc's own server banner style) showing scenario/mode/URL, with
display-width-aware padding so wide characters don't misalign the frame.

…6/UP037/UP045)

Replaces typing.List/Dict/Tuple/Optional/Union with list/dict/tuple/X | None/X | Y in manual_scenarios.py, helpers/editor_scene.py, and helpers/pointer.py (UP006/UP045, no safe autofix available); also picks up ruff --fix's automatic removal of a now-redundant quoted forward-reference in helpers/node.py's Requirement.with_node return type (UP037), safe to auto-fix since from __future__ import annotations is already in effect there.

…st pipeline

Add web-optimized video export to the screencast pipeline

Recorded tests/screencast/output/*.webm files are VP8, sized for
editing, not for embedding on the product website. Add `invoke
screencast-optimize-video` (alias `scov`), which re-encodes each
recording into a muted tests/screencast/output/web/<scenario>.webm
(VP9) + <scenario>.mp4 (H.264, for Safari/older browsers) pair,
capped at 1280px wide without upscaling narrower recordings.

Resolves ffmpeg via PATH with a clear error if it's missing.

…nstead of adding one

Switches the scenario to open Low-Level Requirements and add a Rationale
to the existing LLR-1 (was: creating a new HLR-2 requirement).
editor_scene.reveal_change() now diff-inserts changed lines at their real
position in the file instead of only appending at the end. Terminal/demo
scenes get proper flex+overflow scrolling so typed/appended text
auto-scrolls into view.

…ve path in `strictdoc new` output

Pointer's fake cursor starts hidden (opacity:0), appearing only on first
move/click — scenes with no clicks (terminal/editor) no longer show it
frozen in a corner. run_strictdoc_new now passes a relative path + cwd
so the command's own printed "Location"/"cd" output stays short instead
of leaking pytest's absolute tmp_path.

…video

Navigates to the terminal scene before running the real `strictdoc new`
subprocess — video recording starts as soon as the browser context
exists, so running the subprocess first left about:blank white on
screen for the first second of the recording.

… narrow viewports

min-width: 0 override for the flexbox min-width:auto default, which was
making width: 99% ineffective at narrow widths (flexbug).

…eal fix

Commented out, not deleted — this margin exists so the last anchor in a
document can scroll fully into view, but also causes an unwanted
jump-to-bottom when following any fragment link on a short page.
Disabling here rather than fixing properly (a real fix needs the margin
to apply conditionally, e.g. via a JS-toggled class, not pure CSS).

…nale

Waits for the just-opened edit form's height to stabilize (it keeps
growing briefly right after appearing) before scrolling .main to the
bottom — otherwise the scroll undershoots against still-growing content.

… overflow

… bar