monoid-agent-kernel 0.18.0__tar.gz → 0.19.0__tar.gz

monoid_agent_kernel-0.19.0/.gitattributes

@@ -0,0 +1 @@
+src/monoid_agent_kernel/reference/studio/web/dist/assets/*.js whitespace=-trailing-space

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/.github/workflows/ci.yml

@@ -6,6 +6,28 @@ on:
   pull_request:
 
 jobs:
+  studio-ui:
+    name: Studio UI build
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          cache: npm
+          cache-dependency-path: studio-ui/package-lock.json
+      - name: Install Studio UI dependencies
+        working-directory: studio-ui
+        run: npm ci
+      - name: Type and accessibility check
+        working-directory: studio-ui
+        run: npm run check
+      - name: Build packaged Studio UI
+        working-directory: studio-ui
+        run: npm run build
+      - name: Verify committed assets are current
+        run: git diff --exit-code -- src/monoid_agent_kernel/reference/studio/web/dist
+
   lint:
     runs-on: ubuntu-latest
     steps:

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/.gitignore

@@ -15,11 +15,18 @@ __pycache__/
 .eggs/
 build/
 dist/
+!src/monoid_agent_kernel/reference/studio/web/dist/
+!src/monoid_agent_kernel/reference/studio/web/dist/**
 .pytest_cache/
 .hypothesis/
 .mypy_cache/
 .ruff_cache/
 
+# Studio frontend build workspace
+node_modules/
+studio-ui/coverage/
+studio-ui/.vite/
+
 # Virtual environments
 .venv/
 venv/

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/CHANGELOG.md

@@ -7,6 +7,22 @@ out in commit messages and here.
 
 ## [Unreleased]
 
+## [0.19.0] - 2026-07-13
+
+### Added
+
+- Rebuilt Agent Studio as a packaged Svelte 5, TypeScript, Vite, and Tailwind CSS application with
+  responsive profile, run-control, approval, live-config, change-review, and request-preview
+  surfaces. Released wheels and source distributions carry the compiled UI and require no Node.js
+  runtime.
+- Added explicit Studio pause, resume, and failed-turn retry BFF controls plus a versioned,
+  runtime-resolved `ModelRequest` preview payload.
+
+### Changed
+
+- Moved Studio frontend authoring dependencies into `studio-ui/` and added a deterministic CI build
+  check that keeps the committed Python-package assets synchronized with the Svelte source.
+
 ## [0.18.0] - 2026-07-12
 
 - Added an experimental optional DBOS Reference activation-recovery profile. Its finite,

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: monoid-agent-kernel
-Version: 0.18.0
+Version: 0.19.0
 Summary: Lightweight durable agent kernel for product integrations
 Project-URL: Homepage, https://github.com/hoonseokyoon/monoid-agent-kernel
 Project-URL: Repository, https://github.com/hoonseokyoon/monoid-agent-kernel
@@ -75,18 +75,19 @@ then follow the [Documentation map](#documentation) to the path for your role.
 
 The bundled **Agent Studio** reference app (`monoid studio serve`) drives the kernel
 through its Python API behind a single-page UI. A profile chooses the model, reasoning
-level, prompt instructions, and tool surface; each profile keeps its own chat history.
+level, prompt instructions, and capability surface; each profile keeps its own chat history.
 
-![Agent Studio: a Visual Analyst profile reads sales data, writes insights, and generates an annotated revenue chart](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v016-main.png)
+![Agent Studio: a Visual Analyst reads sales data, pauses for approval, and previews an annotated revenue chart](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v019-main.jpg)
 
-*A real Studio run: the agent reads `sales.csv`, writes `INSIGHTS.md`, generates an
-annotated `revenue_trend.svg`, and previews the artifact directly in the workspace panel.*
+*A real Studio run: the agent reads `sales.csv`, pauses for human approval, and stages
+`INSIGHTS.md` and `revenue_trend.svg` for file-by-file review. The user applies the approved
+files through Studio and previews the chart directly in the workspace panel.*
 
-![Agent Studio profile builder showing the exact system prompt and tool schema preview](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v016-profile-builder.png)
+![Agent Profile Builder with capability controls and the exact ModelRequest preview](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v019-profile-builder.jpg)
 
-*The profile editor shows the exact first-turn model request boundary: system prompt,
-tool schemas, model settings, and preview notes. Users can edit the profile on the left
-and see what the model will receive on the right.*
+*The Profile Builder places model, reasoning, and capability controls beside the
+backend-resolved first-turn `ModelRequest`: system prompt, exact tool schemas, model
+settings, and boundary notes.*
 
 ## Architecture: Contract / Conformance Test / Core Helper Kit / Reference
 
@@ -119,6 +120,10 @@ For the dynamic binding-based tool surface, see
 pip install monoid-agent-kernel
 ```
 
+The Studio UI ships as precompiled static assets. Running Studio does not install or require
+Node.js, Svelte, Vite, or Tailwind CSS; those tools are confined to frontend development and the
+release build.
+
 The default install runs Studio with complete one-shot gateway turns. Add the async HTTP extra
 to render live token deltas in Studio:
 

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/README.md

@@ -24,18 +24,19 @@ then follow the [Documentation map](#documentation) to the path for your role.
 
 The bundled **Agent Studio** reference app (`monoid studio serve`) drives the kernel
 through its Python API behind a single-page UI. A profile chooses the model, reasoning
-level, prompt instructions, and tool surface; each profile keeps its own chat history.
+level, prompt instructions, and capability surface; each profile keeps its own chat history.
 
-![Agent Studio: a Visual Analyst profile reads sales data, writes insights, and generates an annotated revenue chart](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v016-main.png)
+![Agent Studio: a Visual Analyst reads sales data, pauses for approval, and previews an annotated revenue chart](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v019-main.jpg)
 
-*A real Studio run: the agent reads `sales.csv`, writes `INSIGHTS.md`, generates an
-annotated `revenue_trend.svg`, and previews the artifact directly in the workspace panel.*
+*A real Studio run: the agent reads `sales.csv`, pauses for human approval, and stages
+`INSIGHTS.md` and `revenue_trend.svg` for file-by-file review. The user applies the approved
+files through Studio and previews the chart directly in the workspace panel.*
 
-![Agent Studio profile builder showing the exact system prompt and tool schema preview](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v016-profile-builder.png)
+![Agent Profile Builder with capability controls and the exact ModelRequest preview](https://raw.githubusercontent.com/hoonseokyoon/monoid-agent-kernel/main/docs/img/studio-v019-profile-builder.jpg)
 
-*The profile editor shows the exact first-turn model request boundary: system prompt,
-tool schemas, model settings, and preview notes. Users can edit the profile on the left
-and see what the model will receive on the right.*
+*The Profile Builder places model, reasoning, and capability controls beside the
+backend-resolved first-turn `ModelRequest`: system prompt, exact tool schemas, model
+settings, and boundary notes.*
 
 ## Architecture: Contract / Conformance Test / Core Helper Kit / Reference
 
@@ -68,6 +69,10 @@ For the dynamic binding-based tool surface, see
 pip install monoid-agent-kernel
 ```
 
+The Studio UI ships as precompiled static assets. Running Studio does not install or require
+Node.js, Svelte, Vite, or Tailwind CSS; those tools are confined to frontend development and the
+release build.
+
 The default install runs Studio with complete one-shot gateway turns. Add the async HTTP extra
 to render live token deltas in Studio:
 

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "monoid-agent-kernel"
-version = "0.18.0"
+version = "0.19.0"
 description = "Lightweight durable agent kernel for product integrations"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -96,8 +96,10 @@ exclude = ["**/DX_NOTES.md"]
 force-include = { "src/monoid_agent_kernel/py.typed" = "monoid_agent_kernel/py.typed", "src/native_agent_runner/py.typed" = "native_agent_runner/py.typed" }
 
 [tool.hatch.build.targets.sdist]
-# Keep internal journals and workspace-local release scratch data out of source distributions.
-exclude = ["**/DX_NOTES.md", "/.tmp"]
+# Keep internal journals, workspace-local release scratch data, and the Node-only Studio authoring
+# project out of source distributions. The compiled Studio assets live under the Python package and
+# remain available to wheel/sdist users without Node.js.
+exclude = ["**/DX_NOTES.md", "/.tmp", "/studio-ui"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests"]

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/_version.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from importlib.metadata import PackageNotFoundError, version
 
 PACKAGE_NAME = "monoid-agent-kernel"
-FALLBACK_VERSION = "0.18.0"
+FALLBACK_VERSION = "0.19.0"
 
 
 def package_version() -> str:

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/loop.py

@@ -3495,8 +3495,7 @@ class AgentLoop:
         turn_id: str | None = None,
         parent_id: str | None = None,
     ) -> None:
-        diff_text = context.workspace.diff_patch()
-        diff_path = recorder.write_diff(diff_text)
+        diff_text, diff_path, proposal_payload = recorder.write_proposal_revision(context.workspace)
         recorder.emit(
             "workspace.diff.updated",
             turn_id=turn_id,
@@ -3504,10 +3503,12 @@ class AgentLoop:
             data={
                 "path": str(diff_path.relative_to(recorder.run_dir)),
                 "bytes": len(diff_text.encode("utf-8")),
-                "changed_paths": [public_path(path, self.permission_policy) for path in context.workspace.changed_paths()],
+                "changed_paths": [
+                    public_path(path, self.permission_policy)
+                    for path in context.workspace.changed_paths()
+                ],
             },
         )
-        proposal_payload = recorder.write_proposal_snapshot(context.workspace, diff_path)
         recorder.emit(
             "workspace.proposal.updated",
             turn_id=turn_id,

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/loop_phases.py

@@ -477,8 +477,7 @@ class LoopFinalizer:
         recorder = res.recorder
         workspace = res.workspace
         context.job_manager.cancel_all()
-        diff_path = recorder.write_diff(workspace.diff_patch())
-        proposal_payload = recorder.write_proposal_snapshot(workspace, diff_path)
+        _diff_text, diff_path, proposal_payload = recorder.write_proposal_revision(workspace)
         metrics = self.build_metrics(state, res)
         recorder.write_metrics(metrics)
         recorder.emit(
@@ -535,8 +534,7 @@ class LoopFinalizer:
         loop = self._loop
         recorder = res.recorder
         workspace = res.workspace
-        diff_path = recorder.write_diff(workspace.diff_patch())
-        proposal_payload = recorder.write_proposal_snapshot(workspace, diff_path)
+        _diff_text, diff_path, proposal_payload = recorder.write_proposal_revision(workspace)
         metrics = self.build_metrics(state, res)
         recorder.write_metrics(metrics)
         recorder.emit(

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/recorder.py

@@ -3,7 +3,10 @@ from __future__ import annotations
 import json
 import logging
 import sys
+import threading
 import time
+from collections.abc import Iterator
+from contextlib import contextmanager
 from dataclasses import dataclass, field
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, TextIO
@@ -19,6 +22,24 @@ if TYPE_CHECKING:
     from monoid_agent_kernel.core.workspace import ChangedEntry, Workspace
 
 _LOGGER = logging.getLogger("monoid_agent_kernel.recorder")
+_PROPOSAL_LOCKS_GUARD = threading.Lock()
+_PROPOSAL_LOCKS: dict[str, threading.RLock] = {}
+
+
+@contextmanager
+def proposal_snapshot_lock(run_dir: Path) -> Iterator[None]:
+    """Serialize one run's proposal snapshot writes and live-directory consumers.
+
+    Proposal files form one logical revision even though they occupy several paths. The reference
+    backend and Studio use this same in-process lock while reading, exporting, approving, or
+    applying them, so no consumer can observe the writer between ``diff.patch`` and the final
+    atomic ``proposal.json`` replace.
+    """
+    key = str(run_dir.resolve())
+    with _PROPOSAL_LOCKS_GUARD:
+        lock = _PROPOSAL_LOCKS.setdefault(key, threading.RLock())
+    with lock:
+        yield
 
 
 @dataclass
@@ -255,6 +276,14 @@ class AgentRecorder:
         diff_path.write_text(diff_text, encoding="utf-8")
         return diff_path
 
+    def write_proposal_revision(self, workspace: Workspace) -> tuple[str, Path, dict[str, Any]]:
+        """Persist one internally consistent diff and proposal snapshot revision."""
+        with proposal_snapshot_lock(self.run_dir):
+            diff_text = workspace.diff_patch()
+            diff_path = self.write_diff(diff_text)
+            proposal_payload = self.write_proposal_snapshot(workspace, diff_path)
+        return diff_text, diff_path, proposal_payload
+
     def write_manifest(self, manifest: RunManifest) -> Path:
         manifest_path = self.run_dir / "manifest.json"
         write_json_atomic(manifest_path, manifest.to_json())

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/reference/backend/proposal.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import hashlib
 import re
 from collections.abc import Callable
 from dataclasses import dataclass
@@ -16,6 +17,7 @@ from monoid_agent_kernel.core.packages import (
 )
 from monoid_agent_kernel.core.proposal_file import ProposalFileError, read_proposal_file_payload
 from monoid_agent_kernel.errors import PermissionDenied
+from monoid_agent_kernel.recorder import proposal_snapshot_lock
 from monoid_agent_kernel.reference.backend.ports import RunRecordPort
 from monoid_agent_kernel.reference.backend.proposal_reader import read_proposal_snapshot
 from monoid_agent_kernel.reference.backend.run_state import (
@@ -44,7 +46,8 @@ class ProposalService:
     def proposal(self, run_id: str, token: str) -> dict[str, Any]:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
-        payload = read_proposal_snapshot(record)
+        with proposal_snapshot_lock(record.run_dir):
+            payload = read_proposal_snapshot(record)
         if payload is None:
             return {
                 "run_id": record.run_id,
@@ -65,23 +68,31 @@ class ProposalService:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
         diff_path = record.run_dir / "diff.patch"
-        diff = diff_path.read_text(encoding="utf-8") if diff_path.exists() else ""
-        return {"run_id": run_id, "ready": diff_path.exists(), "diff": diff}
+        with proposal_snapshot_lock(record.run_dir):
+            ready = diff_path.exists()
+            data = diff_path.read_bytes() if ready else b""
+        return {
+            "run_id": run_id,
+            "ready": ready,
+            "diff": data.decode("utf-8"),
+            "diff_sha256": hashlib.sha256(data).hexdigest(),
+        }
 
     def proposal_file(self, run_id: str, token: str, path: str) -> dict[str, Any]:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
-        proposal = read_proposal_snapshot(record)
-        if proposal is None:
-            raise ValueError("proposal snapshot is not ready")
-        try:
-            file_payload = read_proposal_file_payload(record.run_dir, proposal, path)
-        except ProposalFileError as exc:
-            if exc.reason in {"not_found", "snapshot_missing"}:
-                raise KeyError(str(exc)) from exc
-            if exc.reason == "escapes_run_dir":
-                raise PermissionDenied(str(exc)) from exc
-            raise ValueError(str(exc)) from exc
+        with proposal_snapshot_lock(record.run_dir):
+            proposal = read_proposal_snapshot(record)
+            if proposal is None:
+                raise ValueError("proposal snapshot is not ready")
+            try:
+                file_payload = read_proposal_file_payload(record.run_dir, proposal, path)
+            except ProposalFileError as exc:
+                if exc.reason in {"not_found", "snapshot_missing"}:
+                    raise KeyError(str(exc)) from exc
+                if exc.reason == "escapes_run_dir":
+                    raise PermissionDenied(str(exc)) from exc
+                raise ValueError(str(exc)) from exc
         return {
             "run_id": record.run_id,
             "tenant_id": record.tenant_id,
@@ -93,8 +104,9 @@ class ProposalService:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
         output = record.run_dir / "proposal.tar"
-        payload = export_package(record.run_dir, output)
-        tar_bytes = output.read_bytes()
+        with proposal_snapshot_lock(record.run_dir):
+            payload = export_package(record.run_dir, output)
+            tar_bytes = output.read_bytes()
         checkpoint_store = self._checkpoint_store()
         digest = checkpoint_store.put_blob(run_id, tar_bytes)
         self._context.emit_backend_event(
@@ -141,13 +153,14 @@ class ProposalService:
     ) -> dict[str, Any]:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
-        approval = create_approval(
-            record.run_dir,
-            approver_id=approver_id,
-            approved_paths=approved_paths or None,
-            note=note,
-        )
-        write_approval(record.run_dir / "approval.json", approval)
+        with proposal_snapshot_lock(record.run_dir):
+            approval = create_approval(
+                record.run_dir,
+                approver_id=approver_id,
+                approved_paths=approved_paths or None,
+                note=note,
+            )
+            write_approval(record.run_dir / "approval.json", approval)
         self._context.emit_backend_event(
             run_id,
             "proposal.approved",
@@ -165,13 +178,14 @@ class ProposalService:
     ) -> dict[str, Any]:
         self._context.authorize_run(run_id, token)
         record = self._context.record(run_id)
-        approval = create_approval(
-            record.run_dir,
-            approver_id=approver_id,
-            decision="rejected",
-            note=reason,
-        )
-        write_approval(record.run_dir / "approval.json", approval)
+        with proposal_snapshot_lock(record.run_dir):
+            approval = create_approval(
+                record.run_dir,
+                approver_id=approver_id,
+                decision="rejected",
+                note=reason,
+            )
+            write_approval(record.run_dir / "approval.json", approval)
         self._context.emit_backend_event(
             run_id,
             "proposal.rejected",
@@ -197,8 +211,9 @@ class ProposalService:
             raise PermissionDenied(f"apply target is outside allowed roots: {target}")
         record = self._context.record(run_id)
         approval = approval_path or (record.run_dir / "approval.json")
-        result = apply_package(record.run_dir, approval=approval, target=target, dry_run=dry_run)
-        write_apply_result(record.run_dir / "apply-result.json", result)
+        with proposal_snapshot_lock(record.run_dir):
+            result = apply_package(record.run_dir, approval=approval, target=target, dry_run=dry_run)
+            write_apply_result(record.run_dir / "apply-result.json", result)
         event_type = "proposal.conflict" if result.status == "conflict" else "proposal.applied"
         self._context.emit_backend_event(
             run_id,

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/reference/backend/recovery.py

@@ -22,7 +22,11 @@ from monoid_agent_kernel.core.durable_metadata import (
     DurableMetadataCommitter,
     validate_recovery_metadata,
 )
-from monoid_agent_kernel.core.result import AgentRunResult, Suspension
+from monoid_agent_kernel.core.result import (
+    AgentRunResult,
+    Suspension,
+    suspension_from_checkpoint_payload,
+)
 from monoid_agent_kernel.identifiers import namespaced_id
 from monoid_agent_kernel.reference.backend.ports import (
     DriveOpenSessionPort,
@@ -209,15 +213,24 @@ class RecoveryService:
         record.seen_inbox_ids = set(checkpoint.inbox_seen_ids)
         for message in checkpoint.queued_messages:
             self._context.call_soon(record.message_queue.put_nowait, message)
-        self._context.spawn(self.run_recovered(run_id, request, loop))
+        restored_suspension = (
+            suspension_from_checkpoint_payload(checkpoint.last_suspension)
+            if checkpoint.last_suspension is not None
+            else Suspension(reason="settled", status="completed")
+        )
+        self._context.spawn(self.run_recovered(run_id, request, loop, restored_suspension))
 
-    async def run_recovered(self, run_id: str, request: RunRequestPort, loop: LoopPort) -> None:
+    async def run_recovered(
+        self,
+        run_id: str,
+        request: RunRequestPort,
+        loop: LoopPort,
+        suspension: Suspension,
+    ) -> None:
         await self._context.acquire_run_slot()
         try:
             if loop.has_pending_tasks():
                 suspension = Suspension(reason="awaiting_tasks", status="running", has_external=True)
-            else:
-                suspension = Suspension(reason="settled", status="completed")
             record = self._context.record(run_id)
             result = await self._context.drive_open_session(
                 record,

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/reference/studio/README.md

@@ -22,6 +22,33 @@ monoid studio accept         # deterministic offline acceptance check, emits JSO
 Run `studio doctor` first if anything looks off — it turns late, cryptic setup failures into an
 upfront pass/fail checklist with remediation.
 
+## Frontend development
+
+Studio's browser UI is a Svelte 5 + TypeScript application under `studio-ui/`. Tailwind CSS v4
+maps the semantic Studio tokens to utilities. Node.js is needed only when authoring or rebuilding
+the UI; released Python packages include the compiled assets.
+
+```bash
+# terminal 1: BFF and agent runtime
+monoid studio serve --port 8799
+
+# terminal 2: Vite dev server with /api proxied to Studio
+cd studio-ui
+npm ci
+npm run dev
+```
+
+Before committing frontend changes, rebuild the packaged assets and run the Svelte checks:
+
+```bash
+cd studio-ui
+npm run check
+npm run build
+```
+
+The build writes to `src/monoid_agent_kernel/reference/studio/web/dist/`. Commit that directory so
+wheel and sdist users can run Studio without Node.js.
+
 ### Flags & defaults
 
 | Flag | Default | Meaning |
@@ -59,6 +86,19 @@ at another file, or `--no-env-file` to rely only on the process environment.
   files, jobs, proposed changes, and file previews. Trace shows the live event tree. Live Config
   exposes model, reasoning, OTel, and capability toggles.
 
+## Review flow
+
+Write-capable profiles stage workspace changes as a proposal. Studio presents human approval,
+per-file review decisions, package export, and apply as separate workflow steps.
+
+![Review Changes with two approved files and a proposed SVG preview](../../../../docs/img/studio-v019-review-changes.jpg)
+
+*Review Changes previews files from the proposal snapshot before apply. Each file can be approved or
+excluded independently; **Apply approved** writes only the approved paths into the workspace.*
+
+**Export package** builds a digest-addressed TAR from the complete current proposal. Per-file review
+decisions control apply and do not filter the exported package.
+
 Profile state is lightweight Studio metadata. The `studio-profiles.json` sidecar under the run root
 stores custom profile presets and maps run ids to profile ids so scoped history survives a restart.
 Profiles are Studio-owned presets; the kernel contract stays unchanged.

{monoid_agent_kernel-0.18.0 → monoid_agent_kernel-0.19.0}/src/monoid_agent_kernel/reference/studio/cli.py

@@ -73,9 +73,15 @@ def run_acceptance(
         base_url = server.start()
         check("healthz", _http_json(f"{base_url}/healthz").get("ok") is True)
         index_html = _http_text(f"{base_url}/")
-        check("index-static-hooks", "data-testid=\"studio-shell\"" in index_html)
+        check(
+            "index-static-shell",
+            '<div id="app"></div>' in index_html and "/assets/" in index_html,
+        )
         settings_html = _http_text(f"{base_url}/settings")
-        check("settings-static-hooks", "data-testid=\"settings-popup\"" in settings_html)
+        check(
+            "settings-static-shell",
+            '<div id="app"></div>' in settings_html and "/assets/" in settings_html,
+        )
         cfg = _http_json(f"{base_url}/api/config")
         check("config-route", cfg.get("offline") is True and cfg.get("provider") == "offline")
         settings = _http_json(f"{base_url}/api/settings")
@@ -103,7 +109,8 @@ def run_acceptance(
             if settled:
                 final_text = str((settled[-1].get("data") or {}).get("final_text") or "")
                 state = str(server.run_status(run_id).get("state") or "")
-                break
+                if state != "running":
+                    break
             time.sleep(0.1)
         check("deterministic-chat", bool(final_text), final_text[:120])
         transcript = _http_json(f"{base_url}/api/chat-transcript?run_id={quote(run_id)}") if run_id else {}