quickstart: --app openemr runs the three acts on a real OpenEMR

CAPABILITIES.md

@@ -144,6 +144,7 @@ improvising or retrying when something is off.
 | Capability | Status | Evidence | Where |
 | --- | --- | --- | --- |
 | Compile one or more demonstrations into a deterministic program | Released | Real app (OpenEMR referral, 62 timed runs) | `openadapt-flow record`, `compile`; [packages/flow](packages/flow/README.md) |
+| Try the three acts on a real app: `openadapt quickstart --app openemr` runs the certified OpenEMR referral program in Docker, five referrals at once checked in OpenEMR's record, then a possible duplicate stopped before saving; without Docker or a clone it says so and exits 2 | On main ([#1431](https://github.com/OpenAdaptAI/OpenAdapt/pull/1431)) | Real app (local OpenEMR 8.0.0.3, synthetic patients) | `openadapt quickstart --app openemr`; [docs/quickstart](docs/quickstart/README.md#on-a-real-app---app-openemr) |
 | Classify common local edits without a person (select-all, Enter in a field or dropdown, an in-form tab, an informational dialog's dismiss), only when the recording's network capture shows no write and a runtime check confirms only the expected change | Released ([#1366](https://github.com/OpenAdaptAI/OpenAdapt/pull/1366)) | Synthetic; OpenEMR referral recording re-admitted offline (violations 12 to 1 under `permissive`) | [LOCAL_EDIT_CLASSIFICATION.md](packages/flow/docs/design/LOCAL_EDIT_CLASSIFICATION.md), `compiler/local_edit.py` |
 | Keep the empty search result a person confirmed through (a duplicate check's "No matches were found.") as an exact check after the click that opens the dialog and again right before Confirm, so a run whose dialog lists a match stops before the save; only fixed-vocabulary search-result statements shown once, never text with a name or search term, and a confirmation dialog without one is noted in `compile_notes.json` | On main | Real app frames (an OpenEMR referral recording, compiled offline) | `compiler/empty_result_guard.py` |
 | Mine that check in any language: induction keeps the one line a named dialog adds (not its title or buttons, not a question, no typed value or number) when two or more demonstrations whose typed values all differ show the same line, refuses when the demonstrations disagree on a mined check instead of copying the first one's, and qualification then needs a business-rule run that halts at it | On main | Synthetic (a search dialog recorded in Chromium by the real recorder, German statement, real OCR) | `compiler/empty_result_guard.py`, `compiler/induction.py` |

docs/quickstart/README.md

@@ -10,3 +10,58 @@
 - `timing-2026-10-20.json`: the engine tutorial with `--break-it` (what
   `openadapt quickstart --break-it` ran before) against the new quickstart,
   three rounds each, interleaved, from `scripts/quickstart_timing.py`.
+
+## On a real app: `--app openemr`
+
+```bash
+openadapt quickstart --app openemr
+```
+
+The same three acts on OpenEMR 8.0.0.3, a real open-source EHR, instead of the
+bundled MockMed. The default stays MockMed: it needs no Docker and finishes in
+about 22 seconds.
+
+- **Needs** Docker (Docker Desktop, or Podman with `DOCKER_HOST` pointed at
+  its socket) and a clone of this repository, because the OpenEMR fixture is
+  repository-only:
+
+  ```bash
+  git clone https://github.com/OpenAdaptAI/OpenAdapt && cd OpenAdapt
+  pip install -e . -e packages/flow
+  ```
+
+  Without either, the command says which one is missing in one line and exits
+  with status 2. It never falls back to MockMed.
+- **Setup.** The pinned fixture from `packages/flow/benchmark/openemr_local/`:
+  pinned image digests, verified source files, a hashed SQL baseline, loopback
+  ports 9741 (HTTP) and 9740 (HTTPS) (set `OPENADAPT_QUICKSTART_OPENEMR_HTTP_PORT`
+  and `OPENADAPT_QUICKSTART_OPENEMR_HTTPS_PORT` to move them), and its own Compose project
+  (`openadapt-quickstart-openemr`). Its state stays in
+  `~/.cache/openadapt/quickstart-openemr`. The first run downloads the images
+  if they're missing (about 2.6 GB unpacked), installs OpenEMR, and adds
+  synthetic API clients. Later runs reuse the running copy and restore its
+  baseline.
+- **Act 1, show it once.** The referral as it was demonstrated once in
+  OpenEMR: the recorded actions of the certified referral program, the same
+  program behind the site's OpenEMR Seals
+  (`benchmark/openemr_local/referral_program/`). It was compiled with the
+  shared duplicate rule (`apply_referral_rules`) and qualified under
+  `clinical-write`.
+- **Act 2, run it.** Five synthetic referrals at once, each on its own clerk
+  account and browser. Each runs through the certified run path (the Standard
+  run gate, then the REST record check) and is read back from OpenEMR's
+  database with SQL: "✓ checked in OpenEMR's record".
+- **Act 3, break it.** The quickstart adds an existing chart (Rosa Delgado,
+  born 1958-03-27), then runs a referral for a Rosa Delgado born 1958-03-22.
+  OpenEMR's duplicate check lists the match, and the program stops before
+  saving ("Stopped before saving: possible duplicate"). SQL finds 0 new charts.
+- **Finale.** The same receipt card and Seal. Every quickstart, in either app,
+  ends the receipt with "See real Seals from OpenEMR runs:
+  https://openadapt.ai/seals".
+
+To stop the fixture, run
+`docker compose -p openadapt-quickstart-openemr stop`. The next quickstart
+starts it again.
+
+The real run is covered by `test_real_openemr_quickstart` (opt in with
+`OPENADAPT_REAL_QUICKSTART_OPENEMR=1` on a machine with Docker); CI skips it.

docs/quickstart/quickstart-openemr.tape

@@ -0,0 +1,35 @@
+# Records docs/quickstart/quickstart-openemr.gif with VHS (https://github.com/charmbracelet/vhs, MIT).
+#
+#   PATH="$PWD/.venv/bin:$PATH" vhs docs/quickstart/quickstart-openemr.tape
+#
+# Run it from a clone with Docker running (with Podman, export DOCKER_HOST
+# first; the recording's shell inherits it). The fixture from an earlier run
+# is reused, so this records a warm run. The browsers stay hidden (no display
+# in the recording). Everything else is the real command, start to finish, at
+# real speed.
+
+Output docs/quickstart/quickstart-openemr.gif
+
+Set Shell bash
+Set FontSize 15
+Set Width 1240
+Set Height 1100
+Set Padding 24
+Set Framerate 12
+Set TypingSpeed 45ms
+Set Theme "Builtin Dark"
+
+Hide
+Type "unset DISPLAY WAYLAND_DISPLAY; export SSH_CONNECTION=recording BROWSER=true OPENADAPT_FLOW_SCRUB=off OPENADAPT_SEAL_HOME=$(mktemp -d) PS1='$ '; cd $(mktemp -d) && clear"
+Enter
+Show
+
+Sleep 500ms
+Type "openadapt quickstart --app openemr"
+Sleep 400ms
+Enter
+Wait+Screen@600s /Press Enter/
+Sleep 2s
+Enter
+Wait+Screen@300s /Give your agent/
+Sleep 8s

llms-full.txt

@@ -578,6 +578,7 @@ improvising or retrying when something is off.
 | Capability | Status | Evidence | Where |
 | --- | --- | --- | --- |
 | Compile one or more demonstrations into a deterministic program | Released | Real app (OpenEMR referral, 62 timed runs) | `openadapt-flow record`, `compile`; [packages/flow](https://github.com/OpenAdaptAI/OpenAdapt/blob/main/packages/flow/README.md) |
+| Try the three acts on a real app: `openadapt quickstart --app openemr` runs the certified OpenEMR referral program in Docker, five referrals at once checked in OpenEMR's record, then a possible duplicate stopped before saving; without Docker or a clone it says so and exits 2 | On main ([#1431](https://github.com/OpenAdaptAI/OpenAdapt/pull/1431)) | Real app (local OpenEMR 8.0.0.3, synthetic patients) | `openadapt quickstart --app openemr`; [docs/quickstart](https://github.com/OpenAdaptAI/OpenAdapt/blob/main/docs/quickstart/README.md#on-a-real-app---app-openemr) |
 | Classify common local edits without a person (select-all, Enter in a field or dropdown, an in-form tab, an informational dialog's dismiss), only when the recording's network capture shows no write and a runtime check confirms only the expected change | Released ([#1366](https://github.com/OpenAdaptAI/OpenAdapt/pull/1366)) | Synthetic; OpenEMR referral recording re-admitted offline (violations 12 to 1 under `permissive`) | [LOCAL_EDIT_CLASSIFICATION.md](https://github.com/OpenAdaptAI/OpenAdapt/blob/main/packages/flow/docs/design/LOCAL_EDIT_CLASSIFICATION.md), `compiler/local_edit.py` |
 | Keep the empty search result a person confirmed through (a duplicate check's "No matches were found.") as an exact check after the click that opens the dialog and again right before Confirm, so a run whose dialog lists a match stops before the save; only fixed-vocabulary search-result statements shown once, never text with a name or search term, and a confirmation dialog without one is noted in `compile_notes.json` | On main | Real app frames (an OpenEMR referral recording, compiled offline) | `compiler/empty_result_guard.py` |
 | Mine that check in any language: induction keeps the one line a named dialog adds (not its title or buttons, not a question, no typed value or number) when two or more demonstrations whose typed values all differ show the same line, refuses when the demonstrations disagree on a mined check instead of copying the first one's, and qualification then needs a business-rule run that halts at it | On main | Synthetic (a search dialog recorded in Chromium by the real recorder, German statement, real OCR) | `compiler/empty_result_guard.py`, `compiler/induction.py` |

openadapt/cli.py

@@ -566,6 +566,7 @@ def _is_externally_managed_error(error: BaseException) -> bool:
 _QUICKSTART_OPTIONS = frozenset(
     {
         "--out",
+        "--app",
         "--headless",
         "--no-break",
         "--verbose",
@@ -654,6 +655,17 @@ def _quickstart_root(out: Optional[Path]) -> Path:
         f"{_DEFAULT_QUICKSTART_DIR}, suffixed -2, -3, ... when taken."
     ),
 )
+@click.option(
+    "--app",
+    type=click.Choice(["mockmed", "openemr"]),
+    default="mockmed",
+    show_default=True,
+    help=(
+        "The app to run it in. mockmed is bundled (about 22 seconds, no Docker). "
+        "openemr runs OpenEMR 8.0.0.3 in Docker from a clone of the repository; "
+        "the first run downloads and installs it, later runs reuse it."
+    ),
+)
 @click.option(
     "--headless",
     is_flag=True,
@@ -676,6 +688,7 @@ def _quickstart_root(out: Optional[Path]) -> Path:
 def quickstart(
     command_ctx: click.Context,
     out: Optional[Path],
+    app: str,
     headless: bool,
     no_break: bool,
     verbose: bool,
@@ -690,14 +703,21 @@ def quickstart(
     screen says saved, the database has nothing, and OpenAdapt stops. The
     program makes no AI calls. Nothing in the output directory is overwritten.
 
+    With --app openemr, the same three acts run on OpenEMR 8.0.0.3 in Docker,
+    with the certified referral program: referrals registered at once and
+    checked in OpenEMR's record, then a referral that matches an existing
+    chart, where the program stops before saving. It runs from a clone of the
+    repository and needs Docker; without either it says so and exits 2.
+
     Other flags of the engine tutorial (for example --guided or
     --interactive-record) run that tutorial instead, unchanged.
     """
     _require_supported_python()
     _refuse_unknown_tutorial_options(command_ctx.args)
+    openemr = _openemr_checkout(command_ctx.args) if app == "openemr" else None
     root = _quickstart_root(out)
 
-    if command_ctx.args or not _flow_runs_quickstart():
+    if openemr is None and (command_ctx.args or not _flow_runs_quickstart()):
         # Engine-owned flags (--guided, --interactive-record, and future
         # engine additions) run the engine's own tutorial, as before. So does
         # a Flow older than the pieces the three-act quickstart uses.
@@ -711,7 +731,7 @@ def quickstart(
     steps = next_steps(
         _hint,
         _shell_quote,
-        _agent_serves_tutorial(),
+        openemr is None and _agent_serves_tutorial(),
         keep=_keep_command(),
         shadow=_shadowing_openadapt if _under_uvx() else None,
     )
@@ -723,6 +743,7 @@ def quickstart(
                 break_it=not no_break,
                 verbose=verbose,
                 steps=steps,
+                openemr=openemr,
             )
     except KeyboardInterrupt:
         click.echo(f"\nStopped. Anything written so far is in {root}", err=True)
@@ -731,7 +752,11 @@ def quickstart(
         if _is_externally_managed_error(error):
             _echo_python_remedy()
             raise click.exceptions.Exit(1) from error
-        if type(error).__name__ in ("TutorialError", "DurableAuthorityBusy"):
+        if type(error).__name__ in (
+            "TutorialError",
+            "DurableAuthorityBusy",
+            "FixtureError",
+        ):
             raise click.ClickException(
                 f"The quickstart stopped: {error}\nAnything it wrote is in {root}"
             ) from error
@@ -742,6 +767,31 @@ def quickstart(
         raise click.exceptions.Exit(1)
 
 
+def _openemr_checkout(extra: list[str]) -> Path:
+    """The clone's packages/flow for ``--app openemr``, or exit 2 saying why.
+
+    Each missing piece is one line on stderr. Nothing falls back to MockMed.
+    """
+    from openadapt.quickstart.openemr import (
+        checkout_problem,
+        docker_problem,
+        find_flow_checkout,
+    )
+
+    if extra:
+        raise click.UsageError(
+            f"--app openemr runs the three-act quickstart only; {extra[0]} "
+            "belongs to the engine's MockMed tutorial."
+        )
+    problem = docker_problem()
+    flow = find_flow_checkout() if problem is None else None
+    problem = problem or checkout_problem(flow)
+    if problem is not None or flow is None:
+        click.echo(problem, err=True)
+        raise click.exceptions.Exit(2)
+    return flow
+
+
 # The oldest Flow with every piece the three-act quickstart uses (shared
 # browser, record checks bound to run inputs, the side-by-side page).
 _QUICKSTART_FLOW_MINIMUM = (2, 2, 0)

openadapt/quickstart/__init__.py

@@ -11,8 +11,9 @@
 
 import sys
 from pathlib import Path
-from typing import IO, Callable, Optional
+from typing import IO, Any, Callable, Optional
 
+from openadapt.quickstart.scene import MOCKMED
 from openadapt.quickstart.story import (
     YOUR_APP_URL,
     NextSteps,
@@ -71,21 +72,32 @@ def run_quickstart(
     steps: Callable[[Outcome], NextSteps],
     stream: Optional[IO[str]] = None,
     interactive: Optional[bool] = None,
+    openemr: Optional[Path] = None,
 ) -> Outcome:
-    """Run the quickstart into ``root`` and print it to ``stream``."""
+    """Run the quickstart into ``root`` and print it to ``stream``.
+
+    ``openemr`` is a clone's ``packages/flow`` folder: the three acts then run
+    on OpenEMR in Docker instead of the bundled MockMed.
+    """
 
     from openadapt.quickstart.engine import FlowEngine, display_available, open_page
     from openadapt.quickstart.view import make_view
 
     stream = stream or sys.stdout
     if interactive is None:
         interactive = bool(sys.stdin.isatty() and getattr(stream, "isatty", bool)())
-    # A person watching gets a short pause between demonstrated actions; a
-    # hidden browser runs at full speed.
-    engine = FlowEngine(root, visible=visible, pace_s=0.3 if visible else 0.0)
+    engine: Any
+    if openemr is not None:
+        from openadapt.quickstart.openemr import OpenEMREngine
+
+        engine = OpenEMREngine(root, visible=visible, flow=openemr)
+    else:
+        # A person watching gets a short pause between demonstrated actions;
+        # a hidden browser runs at full speed.
+        engine = FlowEngine(root, visible=visible, pace_s=0.3 if visible else 0.0)
     return run_story(
         engine,
-        make_view(stream),
+        make_view(stream, scene=getattr(engine, "scene", MOCKMED)),
         next_steps=steps,
         break_it=break_it,
         interactive=interactive,

openadapt/quickstart/engine.py

@@ -25,6 +25,8 @@
 from pathlib import Path
 from typing import Any, Callable, Optional, Sequence
 
+from openadapt.quickstart.scene import MOCKMED
+
 #: Synthetic notes for the five runs. None of them is clinical advice.
 RUN_NOTES = (
     "Follow-up in two weeks",
@@ -95,6 +97,12 @@ class BreakFacts:
     halt_reason: str
     model_calls: int
     seconds: float
+    #: True only when the run's own evidence shows it stopped at the check
+    #: this act sets off (the record check refuting the save, or the app's
+    #: own check the program waits on). A stop for any other reason, or one
+    #: whose cause the evidence doesn't show, stays False, and the views then
+    #: describe a generic stop instead of the demonstration.
+    stop_attributed: bool = False
     result: Any = field(default=None, repr=False)
 
     @property
@@ -346,6 +354,10 @@ def _sha256(path: Path) -> Optional[str]:
 class FlowEngine:
     """The bundled tutorial, one stage per call, inside one browser process."""
 
+    #: Act 2's inputs: one new note per run.
+    inputs = RUN_NOTES
+    scene = MOCKMED
+
     def __init__(self, root: Path, *, visible: bool, pace_s: float = 0.0) -> None:
         self.root = root
         self.visible = visible
@@ -581,6 +593,9 @@ def break_it(self) -> BreakFacts:
             halt_reason=str(result.halt_reason),
             model_calls=int(getattr(report, "model_calls", 0) or 0),
             seconds=seconds,
+            # The record check read the save back and refuted it: the
+            # silent failure this act injects, not some other refusal.
+            stop_attributed=int(result.effects_refuted) > 0,
             result=result,
         )
 
@@ -605,20 +620,7 @@ def seal(self, runs: list[RunFacts]) -> Optional[SealFacts]:
         no Seal and no command to check one.
         """
 
-        try:
-            from openadapt_flow.offline_seal import seal_home, seal_run
-        except ImportError:
-            return None
-        checked = [run for run in runs if run.checked]
-        if not checked:
-            return None
-        key_id = ""
-        for run in checked:
-            # The configured Seal key, or one generated under the Seal home
-            # on first use (the offline-Seal default).
-            seal = seal_run(run.run_dir, home=seal_home(), enqueue=False)
-            key_id = str(seal.key_id)
-        return SealFacts(key_id=key_id, sealed=len(checked))
+        return seal_runs(runs)
 
     def seal_broken(self, broken: BreakFacts) -> bool:
         """Sign the stopped run's report too; True once its Seal is written.
@@ -723,6 +725,25 @@ def write_page(self, first: RunFacts, broken: BreakFacts) -> Optional[Path]:
         return demo_proof.write_demo_page(evidence)
 
 
+def seal_runs(runs: list[RunFacts]) -> Optional[SealFacts]:
+    """Sign each checked run's report with this computer's Seal key."""
+
+    try:
+        from openadapt_flow.offline_seal import seal_home, seal_run
+    except ImportError:
+        return None
+    checked = [run for run in runs if run.checked]
+    if not checked:
+        return None
+    key_id = ""
+    for run in checked:
+        # The configured Seal key, or one generated under the Seal home on
+        # first use (the offline-Seal default).
+        seal = seal_run(run.run_dir, home=seal_home(), enqueue=False)
+        key_id = str(seal.key_id)
+    return SealFacts(key_id=key_id, sealed=len(checked))
+
+
 def return_focus_to_terminal(
     *, visible: bool, platform: Optional[str] = None, environ=None, run=None
 ) -> bool: