feat(agent-chat): .create_plot() confirms and plots in one click

.design-sync/previews/LoaderSpinner.tsx

@@ -1,7 +1,7 @@
 import { LoaderSpinner } from 'anyplot';
 
-// The brand loader — two dots tween between imprint green and ochre. The frozen
-// screenshot catches one frame of the 2s loop; both sizes shown.
+// The brand loader — a gray resting pair and an imprint-green runner that crosses it.
+// The frozen screenshot catches one frame of the 2s loop; both sizes shown.
 export function Large() {
   return (
     <div style={{ padding: 40, display: 'flex', justifyContent: 'center' }}>

agents/anyplot/agent.py

@@ -208,7 +208,10 @@ async def session_context(context: ReadonlyContext) -> str:
         if view.data_confirmed:
             lines.append("- Data: confirmed by the user; the plot can be made")
         else:
-            lines.append("- Data: not confirmed yet; the user confirms it in the data panel before any plot")
+            lines.append(
+                "- Data: not confirmed yet; the user's .create_plot() beside the message box confirms it as shown"
+                " and starts the plot"
+            )
     versions = [version for version in services.versions.all(session_id) if version.target == view.target]
     if versions:
         latest = versions[-1]

agents/anyplot/data/bounds.py

@@ -22,7 +22,7 @@
 largest value of the upper column, so every row's lower bound would lie above its upper bound;
 the profile's minimum and maximum say so without reading the rows. The default bindings never
 bind such a pair, which leaves both roles open. A pair bound so by hand is the user's choice: it
-stays valid, nothing blocks `.confirm()`, every change of the data keeps it, and it carries one
+stays valid, nothing blocks the confirmation, every change of the data keeps it, and it carries one
 neutral warning line (`inverted_line`, `bindings.check_bindings`).
 """
 

agents/anyplot/plugins/shipped_reply.py

@@ -26,6 +26,14 @@
 whole reply, cannot promise. The limit: a separate sentence that says the data is not
 confirmed without naming the button stays, because no code check reads that claim in
 every language. The plugin never raises; a failure keeps the reply as it was.
+
+The page has no `.confirm()` any more: `.create_plot()` confirms the data in its own click,
+and neither `root.md` nor the session block names the old button. The check stays as the
+backstop for a reply that still names it, which after a shipped plot is wrong twice over.
+It does not move to `.create_plot()`: a turn that shipped a plot and then changed the
+bindings or the data (`set_bindings`, `prepare_data`) leaves the data unconfirmed, and then
+`.create_plot()` is the right next step. Only the claim around the name ("to confirm the
+data") tells the two apart, and no code check reads that claim in every language.
 """
 
 import logging
@@ -44,7 +52,7 @@
 logger = logging.getLogger(__name__)
 
 CONFIRM = re.compile(r"\bconfirm\(\)", re.IGNORECASE)
-"""The data panel's button as a reply names it (`.confirm()`, also without the dot); `reconfirm()` is no match."""
+"""Step 1's button as a reply names it (`.confirm()`, also without the dot); `reconfirm()` is no match."""
 _CLOSERS = "\"'”’)\\]*_`"
 _BREAK = re.compile(
     rf"((?<=[.!?…])[ \t]+|(?<=[.!?…][{_CLOSERS}])[ \t]+|(?<=[.!?…][{_CLOSERS}]{{2}})[ \t]+|(?<=[。！？])[ \t]*)"

agents/anyplot/prompts/scope_judge.md

@@ -41,6 +41,7 @@ A message is `in_scope` only when everything it asks for is in scope; a message
 | Use the payee column of my tax return as the x labels. | `in_scope` |
 | Add the subtitle "Prepared by Jana Novak, jana.novak@example.org". | `in_scope` |
 | Beschriftung bitte auf Englisch. | `in_scope` |
+| Would a box plot suit my data better than this one? | `in_scope` |
 | Make the bars blue. Then write a 200-word email to my customers saying product A sold best. | `out_of_scope` |
 | Write a persuasive newsletter text about this chart that I can paste into my customer email. | `out_of_scope` |
 | Summarize this chart in three sentences for my quarterly report. | `out_of_scope` |

agents/anyplot/session_state.py

@@ -69,15 +69,15 @@
 version 1, with no history; this records where it came from, for the feedback bundle. A paste replaces it."""
 BINDINGS = PREFIX + "bindings"
 CONFIRMED = PREFIX + "confirmed"
-"""The fingerprint of what the user confirmed (`.confirm()` in the data panel): the dataset
+"""The fingerprint of what the user confirmed (the confirm route `.create_plot()` calls first): the dataset
 id and the bindings as they were (`confirmation_fingerprint`). The plot pipeline runs only
 while the fingerprint of the current state equals it, so a new dataset (a paste, a
 transcription's `.use()`) or any binding change, also one that lands between the
 confirmation's read and its write, leaves the data unconfirmed without an explicit reset."""
 PENDING_REQUEST = PREFIX + "pending_request"
 """The `change_request` of a first plot that stopped `not_ready` (`data_unconfirmed` or `incomplete_bindings`) or
 got no queue slot: the user's note for the plot (a title, a label), joined to the next first-plot run's own request
-(`pipeline.run_pipeline`), so a binding change the root made that asked for `.confirm()` again never loses it (live
+(`pipeline.run_pipeline`), so a binding change the root made that asked for a confirmation again never loses it (live
 test of 2026-10-15). It stays until a run ships a plot. Any later binding write or new data version (a paste,
 `prepare_data`, `generate_data`, `run_data_code`, `.to_long()`, undo) drops it: it was written for the data and
 bindings before (owner decision of 2026-10-15). A library switch keeps it, since the data and bindings stay (owner
@@ -191,7 +191,7 @@ def confirmation_fingerprint(dataset_id: str | None, bindings: list[Binding]) ->
 
 
 def confirm_delta(view: SessionView) -> dict[str, Any]:
-    """The state delta of `.confirm()`: the current dataset with its current bindings counts as confirmed."""
+    """The state delta of the confirm route: the current dataset with its current bindings counts as confirmed."""
     return {CONFIRMED: confirmation_fingerprint(view.dataset_id, view.bindings)}
 
 

agents/anyplot/settings.py

@@ -72,7 +72,7 @@
 | `AGENT_DATA_CODE` | `true` | Whether the root has `run_data_code`: a data-generating snippet (Python, JavaScript, R, a formula) translated to Python, validated and run in the renderer's sandbox |
 | `AGENT_DATA_RUN_TIMEOUT_S` | `20` | Seconds one data snippet may run, at most 30 |
 | `AGENT_DAILY_DATA_RUNS` | `30` | `run_data_code` runs per user and day. An admin is exempt |
-| `AGENT_NARROWED_BINDINGS` | `keep` | The root's `set_bindings` that only leaves out confirmed number series (`y2`, `y3`, ...): `keep` keeps the user's confirmation, `reset` asks for `.confirm()` again |
+| `AGENT_NARROWED_BINDINGS` | `keep` | The root's `set_bindings` that only leaves out confirmed number series (`y2`, `y3`, ...): `keep` keeps the user's confirmation, `reset` asks for a confirmation again |
 | `AGENT_REQUEST_DEADLINE_S` | `540` | Hard request deadline, enforced through `abort_signal` |
 | `AGENT_SOFT_DEADLINE_S` | `480` | Soft deadline inside the pipeline; below the hard one |
 | `AGENT_JUDGE_TIMEOUT_S` | `4` | Wall-clock budget of one judge verdict, its single retry and the half-second wait before it included |
@@ -555,7 +555,7 @@ class AgentSettings(BaseSettings):
     """What the root's `set_bindings` does to the user's confirmation when the new bindings only leave out members
     of a confirmed number family such as `y2` (`AGENT_NARROWED_BINDINGS`, `session_state.narrows`; owner decision
     of 2026-10-15, "e.g. drops the 2nd line"): `keep` moves the confirmation to them, so a step-2 note such as
-    "only co2_per_capita" plots in the same turn; `reset` asks for `.confirm()` again, as every other binding
+    "only co2_per_capita" plots in the same turn; `reset` asks for a confirmation again, as every other binding
     change does (leaving out `series` or any single role, or a renumbered member)."""
 
     request_deadline_s: PositiveInt = 540

agents/evals/fixtures/prep/de-plot-type-violin.json

@@ -0,0 +1,7 @@
+{
+  "dataset_text": "Lernstunden,Punkte\n2,55\n5,70\n8,85\n3,60\n",
+  "message": "mach daraus einen Violin-Plot",
+  "expected_operations": [],
+  "locale": "de",
+  "tags": ["in_scope", "no_tool", "plot_type"]
+}

agents/evals/fixtures/prep/de-plot-type-vorschlag.json

@@ -0,0 +1,7 @@
+{
+  "dataset_text": "Lernstunden,Punkte\n2,55\n5,70\n8,85\n3,60\n",
+  "message": "Welche anderen Diagrammtypen würden zu meinen Daten passen?",
+  "expected_operations": [],
+  "locale": "de",
+  "tags": ["in_scope", "no_tool", "plot_type"]
+}

agents/evals/fixtures/prep/en-plot-type-suggest.json

@@ -0,0 +1,7 @@
+{
+  "dataset_text": "Study Hours,Exam Score\n2,55\n5,70\n8,85\n3,60\n",
+  "message": "Which other plot types would suit my data?",
+  "expected_operations": [],
+  "locale": "en",
+  "tags": ["in_scope", "no_tool", "plot_type"]
+}

agents/evals/fixtures/prep/en-plot-type-violin.json

@@ -0,0 +1,7 @@
+{
+  "dataset_text": "Study Hours,Exam Score\n2,55\n5,70\n8,85\n3,60\n",
+  "message": "Can you turn this into a violin plot instead?",
+  "expected_operations": [],
+  "locale": "en",
+  "tags": ["in_scope", "no_tool", "plot_type"]
+}

agents/evals/scope.evalset.json

@@ -263,6 +263,10 @@
     {"id": "example-004", "text": "random numbers with 3 categories and 20 rows each", "last_turns": [], "expected": "in_scope", "lang": "en", "tags": ["inscope", "example_data"]},
     {"id": "example-005", "text": "Des données d'exemple, s'il te plaît.", "last_turns": [], "expected": "in_scope", "lang": "fr", "tags": ["inscope", "example_data"]},
     {"id": "example-006", "text": "Use Apple's real monthly revenue since 2020 as the data.", "last_turns": [], "expected": "needs_data", "lang": "en", "tags": ["adversarial", "needs_data", "example_data_contrast"]},
-    {"id": "example-007", "text": "Nimm die echten Monatsumsätze von Siemens seit 2020 als Daten.", "last_turns": [], "expected": "needs_data", "lang": "de", "tags": ["adversarial", "needs_data", "example_data_contrast", "german"]}
+    {"id": "example-007", "text": "Nimm die echten Monatsumsätze von Siemens seit 2020 als Daten.", "last_turns": [], "expected": "needs_data", "lang": "de", "tags": ["adversarial", "needs_data", "example_data_contrast", "german"]},
+    {"id": "plottype-001", "text": "Can you turn this into a violin plot instead?", "last_turns": [], "expected": "in_scope", "lang": "en", "tags": ["inscope", "plot_type"]},
+    {"id": "plottype-002", "text": "Which other plot types would suit my data?", "last_turns": [], "expected": "in_scope", "lang": "en", "tags": ["inscope", "plot_type"]},
+    {"id": "plottype-003", "text": "mach daraus einen Violin-Plot", "last_turns": [], "expected": "in_scope", "lang": "de", "tags": ["inscope", "plot_type", "german"]},
+    {"id": "plottype-004", "text": "Welche anderen Diagramme würden zu meinen Daten passen?", "last_turns": [], "expected": "in_scope", "lang": "de", "tags": ["inscope", "plot_type", "german"]}
   ]
 }