Merge pull request #2397 from nesquena/stage-368

v0.51.75 — Release AY (stage-368: 11-PR safe-lane batch + pytest-loop fix)

CHANGELOG.md

@@ -2,6 +2,38 @@
 
 ## [Unreleased]
 
+## [v0.51.75] — 2026-05-16 — Release AY (stage-368 — 11-PR safe-lane batch — storage + i18n + run-journal parity + attachments + compression sidebar + restart-recovery + text-mode images + tables + settings i18n + German labels)
+
+### Test infrastructure
+
+- Stage-368 maintainer fix — pytest no longer self-loops on the `_schedule_restart` daemon thread. Several existing tests in `tests/test_update_banner_fixes.py` call `api.updates._schedule_restart()`, which spawns a daemon thread that eventually calls `os.execv()`. Those tests monkeypatch `os.execv` for the test scope, but monkeypatch teardown can win the race against the daemon thread, restoring the real `os.execv` before the thread fires it — at which point the daemon re-execs the entire pytest process with the original argv, looking from the outside like pytest hangs at 99 % then restarts the suite from 0 % in an infinite loop. `tests/conftest.py` now installs a permanent no-op wrapper on `os.execv` at module-import time so late-firing daemon threads cannot re-exec pytest. New `tests/test_pytest_execv_guard.py` pins the guard against future regressions.
+
+### Added
+
+- **PR #2377** by @franksong2702 (refs #2283, refs #2363, refs #1925) — Run-journal replay timeline parity checks. After #2283 shipped the first run-journal replay slice and #2363 documented the cross-layer state consistency contract, this PR adds explicit parity assertions over the replayed timeline so divergences between the journal and the visible transcript (Thinking → tool calls → assistant text) surface as test failures instead of silent drift.
+
+### Fixed
+
+- **PR #2391** by @Michaelyklam (fixes #2389) — Reduce browser storage pressure during service-worker updates and over long-running sessions. `static/sw.js` now calls `deleteOldShellCaches()` BEFORE `caches.open(CACHE_NAME)` in the install handler so the new ~2.2 MB shell cache no longer overlaps the old one during a version bump (especially painful on shared-origin quota accounting). A new `_clearSessionViewedCount()` helper plus extended `_clearHandoffStorageForSession()` prune `hermes-session-viewed-counts`, `hermes-session-completion-unread`, and `hermes-session-observed-streaming` on every single-session delete and batch-delete so per-session tracking maps no longer grow unbounded.
+
+- **PR #2387** by @Michaelyklam (fixes #2386) — Guard `localStorage.setItem('hermes-webui-session', ...)` and workspace-panel runtime-state writes with `try { … } catch (_) {}` across `static/boot.js`, `static/sessions.js`, `static/commands.js`, and `static/messages.js`. These convenience writes were previously fatal UI operations on quota-exhausted browsers (especially Firefox public-domain setups where shared quota fills up after a service-worker shell rotation).
+
+- **PR #2368** by @Michaelyklam — Hybridize background profile env routing so background title generation, manual compression, and update-summary workers honor a session's non-default profile. The pure thread-local refactor for #2321 was reverted because `hermes_cli.config.load_config()` still reads `HERMES_HOME` from process env. This PR keeps the thread-local layer for WebUI helpers and adds an `os.environ.update(runtime_env)` mirror under a narrow `_ENV_LOCK` for the worker body, with proper restore of prior values. New test asserts `OPENROUTER_API_KEY` is visible from the worker against a non-default profile.
+
+- **PR #2382** by @Michaelyklam (fixes #2380) — Serve raw chat attachments from the per-session inbox in addition to the session workspace. Chat uploads were intentionally moved out of workspaces into a per-session attachment inbox in an earlier release; the transcript renderer still emits stable `api/file/raw?session_id=...&path=<filename>` URLs, but `_handle_file_raw` only checked `session.workspace` so inbox-backed uploads rendered as broken images. The URL surface is preserved and a session-attachment fallback is added with path-traversal guards intact.
+
+- **PR #2385** by @franksong2702 — Keep fuller compression snapshots reachable in the sidebar. The default behavior hides `pre_compression_snapshot: true` rows so archived compression segments do not duplicate the active continuation. A real long Kanban session exposed a narrower failure: the fuller transcript was still present on disk but remained marked as `pre_compression_snapshot`, so the sidebar surfaced a shorter row and the fuller transcript became unreachable. The fix preserves discoverability without re-introducing duplication in normal cases.
+
+- **PR #2371** by @franksong2702 — Clarify interrupted turn recovery after a WebUI restart. WebUI executes browser-originated agent turns inside the WebUI process; if that process restarts mid-turn, the worker dies with it. Run journal replay can only replay events that were already emitted, so the stale-pending repair path is now annotated and refined to make the post-restart state explicit (interrupted, recoverable, or terminal) instead of leaving the user with a half-rendered turn and no signal.
+
+- **PR #2378** by @Michaelyklam — Strip historical images in text-only mode. Current-turn uploads already respect `agent.image_input_mode: text`, but saved conversation history still passed native `image_url` content parts back into later provider calls, breaking text-only providers on replayed turns. `_sanitize_messages_for_api()` gains a `cfg=` keyword argument so the API-history sanitizer can strip historical native image parts when the mode is text. Default `cfg=None` preserves prior behavior for callers that don't pass the new argument.
+
+- **PR #2375** by @Michaelyklam — Keep Markdown tables block-level. Pipe tables were already converted to `<table>` markup, but the final paragraph pass did not treat generated tables as block-level output, occasionally wrapping them in `<p>` and breaking the surrounding layout. The fix isolates generated tables and adds `table` to the paragraph-wrap skip list so valid CommonMark tables render predictably.
+
+- **PR #2372** by @mccxj — Settings → Conversation page action buttons now respect locale selection. Pre-fix, the JSON export, MD export, and Copy buttons had hardcoded English labels/titles. Adds `data-i18n` / `data-i18n-title` attributes plus the missing translation keys so non-English locales no longer see English labels stuck in the middle of a translated screen.
+
+- **PR #2381** by @Michaelyklam (fixes #2379) — German relative session-time labels now interpolate the elapsed value instead of rendering the literal `{n}` placeholder in the sidebar/header. The German locale now uses function-valued translations for minutes, hours, and days, matching the other locale bundles.
+
 ## [v0.51.74] — 2026-05-16 — Release AX (stage-367 — 4-PR safe-lane batch — #2362 table-cell spacing + #2363 run-state-consistency RFC + #2365 custom_providers list-format + #2367 settings sidebar i18n)
 
 ### Added

api/models.py

@@ -679,6 +679,20 @@ def _get_profile_home(profile) -> Path:
         return Path(os.environ.get('HERMES_HOME') or '~/.hermes').expanduser()
 
 
+def _interrupted_recovery_marker() -> dict:
+    return {
+        'role': 'assistant',
+        'content': (
+            '**Response interrupted.**\n\n'
+            'The WebUI process restarted before this turn finished. '
+            'The user message above was preserved, but no agent output was recovered.'
+        ),
+        'timestamp': int(time.time()),
+        '_error': True,
+        'type': 'interrupted',
+    }
+
+
 def _apply_core_sync_or_error_marker(
     session,
     core_path,
@@ -745,12 +759,7 @@ def _apply_core_sync_or_error_marker(
         session.pending_user_message = None
         session.pending_attachments = []
         session.pending_started_at = None
-        session.messages.append({
-            'role': 'assistant',
-            'content': '**Previous turn did not complete.**',
-            'timestamp': int(time.time()),
-            '_error': True,
-        })
+        session.messages.append(_interrupted_recovery_marker())
         session.save()
         logger.info(
             "Session %s: recovered pending user turn (messages non-empty), added error marker",
@@ -794,12 +803,7 @@ def _apply_core_sync_or_error_marker(
     session.pending_user_message = None
     session.pending_attachments = []
     session.pending_started_at = None
-    session.messages.append({
-        'role': 'assistant',
-        'content': '**Previous turn did not complete.**',
-        'timestamp': int(time.time()),
-        '_error': True,
-    })
+    session.messages.append(_interrupted_recovery_marker())
     session.save()
     logger.info("Session %s: no core transcript found, added error marker", sid)
     return True
@@ -811,7 +815,7 @@ def _apply_core_sync_or_error_marker(
 # pending_user_message and STREAMS.pop(stream_id). Without this guard, any
 # fast turn (e.g. command approval) that exits the thread before the on-disk
 # pending clear has flushed gets misdiagnosed as a crashed turn, producing a
-# spurious "Previous turn did not complete." marker.
+# spurious "Response interrupted." marker.
 #
 # 30s covers the worst-case post-loop persistence window: LLM finishing a tool
 # batch + lock contention with the checkpoint thread + a multi-MB session.save.
@@ -1012,7 +1016,110 @@ def _hide_from_default_sidebar(session: dict) -> bool:
     """Return True for internal/background sessions hidden from the default list."""
     sid = str(session.get('session_id') or '')
     source = session.get('source_tag') or session.get('source')
-    return bool(session.get('pre_compression_snapshot')) or source == 'cron' or sid.startswith('cron_')
+    if source == 'cron' or sid.startswith('cron_'):
+        return True
+    if bool(session.get('pre_compression_snapshot')):
+        return not bool(session.get('_show_pre_compression_snapshot'))
+    return False
+
+
+def _sidebar_message_count(session: dict) -> int:
+    for key in ('message_count', 'actual_message_count'):
+        try:
+            value = int(session.get(key) or 0)
+        except (TypeError, ValueError):
+            value = 0
+        if value > 0:
+            return value
+    return 0
+
+
+def _sidebar_lineage_root_id(session: dict, sessions_by_id: dict[str, dict]) -> str:
+    sid = str(session.get('session_id') or '')
+    root = sid
+    parent = session.get('parent_session_id')
+    seen = {sid}
+    while parent and parent not in seen and parent in sessions_by_id:
+        root = str(parent)
+        seen.add(root)
+        parent = sessions_by_id.get(root, {}).get('parent_session_id')
+    return root
+
+
+def _has_live_sidebar_state(session: dict) -> bool:
+    return bool(
+        session.get('active_stream_id')
+        or session.get('has_pending_user_message')
+        or session.get('pending_user_message')
+    )
+
+
+def _prefer_fuller_snapshots_for_sidebar(sessions: list[dict]) -> list[dict]:
+    """Expose a hidden snapshot when it is the fuller transcript for a lineage.
+
+    Pre-compression snapshots are normally hidden so archived compression
+    segments do not duplicate the current continuation in the sidebar. If a
+    snapshot row has more messages than the visible continuation for the same
+    lineage, hiding it makes the conversation look truncated. In that case,
+    show the fuller snapshot and suppress the shorter inactive continuation.
+    """
+    sessions_by_id = {
+        str(session.get('session_id')): session
+        for session in sessions
+        if session.get('session_id')
+    }
+    groups: dict[str, list[dict]] = {}
+    for session in sessions:
+        sid = str(session.get('session_id') or '')
+        source = session.get('source_tag') or session.get('source')
+        if source == 'cron' or sid.startswith('cron_'):
+            continue
+        root = _sidebar_lineage_root_id(session, sessions_by_id)
+        groups.setdefault(root, []).append(session)
+
+    snapshot_ids_to_show: set[str] = set()
+    continuation_ids_to_hide: set[str] = set()
+    for group in groups.values():
+        visible = [session for session in group if not session.get('pre_compression_snapshot')]
+        snapshots = [session for session in group if session.get('pre_compression_snapshot')]
+        if not visible or not snapshots:
+            continue
+        if any(_has_live_sidebar_state(session) for session in visible):
+            continue
+
+        best_visible_count = max(_sidebar_message_count(session) for session in visible)
+        best_snapshot = max(
+            snapshots,
+            key=lambda session: (_sidebar_message_count(session), _session_sort_timestamp(session)),
+        )
+        if _sidebar_message_count(best_snapshot) <= best_visible_count:
+            continue
+
+        snapshot_ids_to_show.add(str(best_snapshot.get('session_id')))
+        continuation_ids_to_hide.update(
+            str(session.get('session_id'))
+            for session in visible
+            if session.get('session_id')
+        )
+
+    if not snapshot_ids_to_show and not continuation_ids_to_hide:
+        return sessions
+
+    out = []
+    for session in sessions:
+        sid = str(session.get('session_id') or '')
+        if sid in continuation_ids_to_hide:
+            continue
+        if sid in snapshot_ids_to_show:
+            session = dict(session)
+            session['_show_pre_compression_snapshot'] = True
+        out.append(session)
+    return out
+
+
+def _strip_sidebar_internal_flags(sessions: list[dict]) -> None:
+    for session in sessions:
+        session.pop('_show_pre_compression_snapshot', None)
 
 
 def _active_state_db_path() -> Path:
@@ -1131,7 +1238,9 @@ def all_sessions(diag=None):
                 and not s.get('has_pending_user_message')
                 and not s.get('worktree_path')
             )]
+            result = _prefer_fuller_snapshots_for_sidebar(result)
             result = [s for s in result if not _hide_from_default_sidebar(s)]
+            _strip_sidebar_internal_flags(result)
             # Backfill: sessions created before Sprint 22 have no profile tag.
             # Attribute them to 'default' so the client profile filter works correctly.
             for s in result:
@@ -1167,7 +1276,9 @@ def all_sessions(diag=None):
         and not s.pending_user_message
         and not getattr(s, 'worktree_path', None)
     )]
+    result = _prefer_fuller_snapshots_for_sidebar(result)
     result = [s for s in result if not _hide_from_default_sidebar(s)]
+    _strip_sidebar_internal_flags(result)
     for s in result:
         if not s.get('profile'):
             s['profile'] = 'default'

api/profiles.py

@@ -697,7 +697,8 @@ def profile_env_for_background_worker(
         return
 
     try:
-        # Lazy import avoids a module-load cycle: streaming imports this helper.
+        # Lazy imports avoid a module-load cycle: streaming imports this helper.
+        from api.config import _clear_thread_env, _set_thread_env, _thread_ctx
         from api.streaming import _ENV_LOCK
 
         profile_home_path = Path(get_hermes_home_for_profile(profile))
@@ -712,21 +713,24 @@ def profile_env_for_background_worker(
         yield
         return
 
-    env_keys = set(runtime_env.keys()) | {"HERMES_HOME"}
-    # Stage-360 maintainer fix: narrow the _ENV_LOCK critical section to just
-    # the env mutation (and the env restoration). Pre-fix, this held _ENV_LOCK
-    # for the entire `yield` duration — i.e. the whole background worker's
-    # runtime (title generation, compression, update summary). That caused
-    # _ENV_LOCK to be held for many seconds, blocking ALL other sessions and
-    # surfacing as the QA `test_third_message_completes` timeout. The fix
-    # mirrors the narrow-lock pattern in _run_agent_streaming: acquire briefly
-    # to set env, run worker without holding the lock, reacquire to restore.
-    # See also QA `test_finally_restores_env_with_lock`.
+    thread_env = dict(runtime_env)
+    thread_env["HERMES_HOME"] = str(profile_home_path)
+    # Hybrid profile routing: keep the broad runtime env in WebUI's thread-local
+    # channel for WebUI helpers, and also mirror it into process env for the
+    # worker body because several production Hermes readers still call
+    # os.getenv() directly for provider credentials.  Keep the _ENV_LOCK scope
+    # narrow: serialize only setup/restore, not the whole worker body.
     skill_home_snapshot = None
-    old_env = {}
+    old_runtime_env: dict[str, Optional[str]] = {}
+    old_hermes_home = None
+    had_hermes_home = False
+    previous_thread_env = getattr(_thread_ctx, "env", {}).copy()
     try:
+        _set_thread_env(**thread_env)
         with _ENV_LOCK:
-            old_env = {key: os.environ.get(key) for key in env_keys}
+            old_runtime_env = {key: os.environ.get(key) for key in runtime_env}
+            had_hermes_home = "HERMES_HOME" in os.environ
+            old_hermes_home = os.environ.get("HERMES_HOME")
             skill_home_snapshot = snapshot_skill_home_modules()
             os.environ.update(runtime_env)
             os.environ["HERMES_HOME"] = str(profile_home_path)
@@ -741,12 +745,20 @@ def profile_env_for_background_worker(
                 )
         yield
     finally:
+        if previous_thread_env:
+            _set_thread_env(**previous_thread_env)
+        else:
+            _clear_thread_env()
         with _ENV_LOCK:
-            for key, old_value in old_env.items():
+            for key, old_value in old_runtime_env.items():
                 if old_value is None:
                     os.environ.pop(key, None)
                 else:
                     os.environ[key] = old_value
+            if had_hermes_home:
+                os.environ["HERMES_HOME"] = old_hermes_home or ""
+            else:
+                os.environ.pop("HERMES_HOME", None)
             if skill_home_snapshot is not None:
                 restore_skill_home_modules(skill_home_snapshot)
 

api/routes.py

@@ -6360,10 +6360,34 @@ def _handle_media(handler, parsed):
             or html_inline_ok
         )
     ) else "attachment"
+    # _serve_file_bytes sends Content-Security-Policy when csp is set.
     csp = "sandbox allow-scripts" if html_inline_ok else None
     return _serve_file_bytes(handler, target, mime, disposition, "private, max-age=3600", csp=csp)
 
 
+def _file_raw_target(session, sid: str, rel: str) -> Path | None:
+    """Resolve /api/file/raw paths from the workspace or this session's uploads."""
+    try:
+        target = safe_resolve(Path(session.workspace), rel)
+    except ValueError:
+        target = None
+    if target and target.exists() and target.is_file():
+        return target
+
+    # Chat uploads now live in a per-session attachment inbox outside the
+    # workspace. Keep the public URL stable while scoping fallback lookup to
+    # the requesting session's own attachment directory.
+    try:
+        from api.upload import _session_attachment_dir
+
+        attachment_target = safe_resolve(_session_attachment_dir(sid), rel)
+    except Exception:
+        return None
+    if attachment_target.exists() and attachment_target.is_file():
+        return attachment_target
+    return None
+
+
 def _handle_file_raw(handler, parsed):
     qs = parse_qs(parsed.query)
     sid = qs.get("session_id", [""])[0]
@@ -6375,8 +6399,8 @@ def _handle_file_raw(handler, parsed):
         return bad(handler, "Session not found", 404)
     rel = qs.get("path", [""])[0]
     force_download = qs.get("download", [""])[0] == "1"
-    target = safe_resolve(Path(s.workspace), rel)
-    if not target.exists() or not target.is_file():
+    target = _file_raw_target(s, sid, rel)
+    if target is None:
         return j(handler, {"error": "not found"}, status=404)
     ext = target.suffix.lower()
     mime = MIME_MAP.get(ext, "application/octet-stream")

api/streaming.py

@@ -1831,7 +1831,32 @@ def _maybe_schedule_title_refresh(session, put_event, agent):
     ).start()
 
 
-def _sanitize_messages_for_api(messages):
+def _strip_native_image_parts_from_content(content):
+    """Return provider-safe content with native image parts removed.
+
+    Text-only provider endpoints (for example DeepSeek/OpenAI-compatible text
+    models) reject historical OpenAI-style ``image_url`` parts before the agent
+    can recover.  When WebUI is configured for text-mode image handling, preserve
+    textual content from mixed content arrays and drop only the native image
+    blocks from replayed history.
+    """
+    if not isinstance(content, list):
+        return content
+    clean_parts = []
+    for part in content:
+        if not isinstance(part, dict):
+            continue
+        if part.get('type') == 'image_url' or 'image_url' in part:
+            continue
+        clean_parts.append(copy.deepcopy(part))
+    if not clean_parts:
+        return ''
+    if len(clean_parts) == 1 and clean_parts[0].get('type') == 'text':
+        return str(clean_parts[0].get('text') or '')
+    return clean_parts
+
+
+def _sanitize_messages_for_api(messages, *, cfg: dict = None):
     """Return a deep copy of messages with only API-safe fields.
 
     The webui stores extra metadata on messages (attachments, timestamp, _ts)
@@ -1843,7 +1868,14 @@ def _sanitize_messages_for_api(messages):
     (Mercury-2/Inception, newer OpenAI models) reject histories containing dangling
     tool results with a 400 error: "Message has tool role, but there was no previous
     assistant message with a tool call."
+
+    If ``agent.image_input_mode`` resolves to ``text``, native historical
+    ``image_url`` content parts are stripped too.  Current-turn uploads already
+    respect text mode in ``_build_native_multimodal_message``; this closes the
+    remaining replay gap where an older native image in the saved transcript kept
+    causing 400s on every later text-only turn (#2297).
     """
+    strip_native_images = cfg is not None and _resolve_image_input_mode(cfg) == "text"
     # First pass: collect all tool_call_ids declared by assistant messages.
     # Handles both OpenAI ('id') and Anthropic ('call_id') field names.
     valid_tool_call_ids: set = set()
@@ -1872,6 +1904,8 @@ def _sanitize_messages_for_api(messages):
                 # Orphaned tool result — skip to avoid 400 from strict providers.
                 continue
         sanitized = {k: v for k, v in msg.items() if k in _API_SAFE_MSG_KEYS}
+        if strip_native_images and 'content' in sanitized:
+            sanitized['content'] = _strip_native_image_parts_from_content(sanitized.get('content'))
         if sanitized.get('role'):
             clean.append(sanitized)
     return clean
@@ -3515,7 +3549,7 @@ def _run_agent_streaming(
             result = agent.run_conversation(
                 user_message=user_message,
                 system_message=workspace_system_msg,
-                conversation_history=_sanitize_messages_for_api(_previous_context_messages),
+                conversation_history=_sanitize_messages_for_api(_previous_context_messages, cfg=_cfg),
                 task_id=session_id,
                 persist_user_message=msg_text,
             )
@@ -3726,7 +3760,7 @@ def _run_agent_streaming(
                                 _heal_result = agent.run_conversation(
                                     user_message=user_message,
                                     system_message=workspace_system_msg,
-                                    conversation_history=_sanitize_messages_for_api(_previous_context_messages),
+                                    conversation_history=_sanitize_messages_for_api(_previous_context_messages, cfg=_cfg),
                                     task_id=session_id,
                                     persist_user_message=msg_text,
                                 )
@@ -4505,7 +4539,7 @@ def _run_agent_streaming(
                         _heal_result = _heal_agent.run_conversation(
                             user_message=user_message,
                             system_message=workspace_system_msg,
-                            conversation_history=_sanitize_messages_for_api(_previous_context_messages),
+                            conversation_history=_sanitize_messages_for_api(_previous_context_messages, cfg=_cfg),
                             task_id=session_id,
                             persist_user_message=msg_text,
                         )

docs/rfcs/webui-run-state-consistency-contract.md

@@ -78,7 +78,10 @@ while WebUI still has multiple overlapping state stores.
    assistant just acted.
 5. **Replay is idempotent.** Replaying a run from a cursor must not duplicate
    transcript rows, thinking content, interim assistant text, tool cards, or
-   compression cards.
+   compression cards. Replayed long-task events should enter the same
+   browser-facing timeline renderer as live SSE events so recovery does not
+   downgrade a structured Thinking / progress / tool / compression turn into a
+   separate flattened presentation.
 6. **Compression is not current intent.** Automatic compression summaries and
    reference cards are recovery/handoff material. They must not be treated as a
    new user request, active-turn content, or the default visible explanation for
@@ -102,6 +105,8 @@ context reconstruction, or session metadata:
 - What happens after browser refresh, session switch, SSE reconnect, and WebUI
   restart?
 - Does replay rebuild the same scene without duplicates?
+- Does replay use the same timeline-rendering path as live SSE for thinking,
+  interim assistant text, tool cards, compression cards, and terminal states?
 - Can this change move a session in the sidebar without meaningful user or
   assistant activity?
 - Can automatic compression or recovery text become visible active-turn content?
@@ -147,4 +152,3 @@ The two documents should be read together:
 4. If #1925 introduces a new adapter-backed runtime layer, update this RFC or
    replace it with the accepted implementation contract so these invariants do
    not live only in historical discussion.
-

static/boot.js

@@ -101,7 +101,7 @@ function _setWorkspacePanelMode(mode){
   // Persist open/closed across refreshes (browse/preview → open; closed → closed)
   // Do NOT overwrite the user's "keep open" preference — only track runtime state
   // so that toggleWorkspacePanel(false) from the toolbar doesn't clear the setting.
-  localStorage.setItem('hermes-webui-workspace-panel', open ? 'open' : 'closed');
+  try{localStorage.setItem('hermes-webui-workspace-panel', open ? 'open' : 'closed');}catch(_){}
   layout.classList.toggle('workspace-panel-collapsed',!open);
   if(_isCompactWorkspaceViewport()){
     panel.classList.toggle('mobile-open',open);

static/commands.js

@@ -424,7 +424,7 @@ async function _applyManualCompressionResult(data, focusTopic, visibleCount, com
       S.messages=data.session.messages||[];
       S.toolCalls=data.session.tool_calls||[];
       clearLiveToolCards();
-      localStorage.setItem('hermes-webui-session',S.session.session_id);
+      try{localStorage.setItem('hermes-webui-session',S.session.session_id);}catch(_){}
       if(typeof _setActiveSessionUrl==='function') _setActiveSessionUrl(S.session.session_id);
       syncTopbar();
       renderMessages();

static/i18n.js

@@ -731,6 +731,10 @@ const LOCALES = {
     transcript: 'Transcript',
     download_transcript: 'Download as Markdown',
     import: 'Import',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Export full session as JSON',
+    import_session_json_tooltip: 'Import session from JSON',
+    clear_conversation_btn_tooltip: 'Clear all messages in this conversation',
     // Settings detail
     settings_label_sound: 'Notification sound',
     settings_desc_sound: 'Play a sound when the assistant finishes a response.',
@@ -1913,6 +1917,10 @@ const LOCALES = {
     transcript: 'Trascrizione',
     download_transcript: 'Scarica come Markdown',
     import: 'Importa',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Esporta sessione completa come JSON',
+    import_session_json_tooltip: 'Importa sessione da JSON',
+    clear_conversation_btn_tooltip: 'Cancella tutti i messaggi in questa conversazione',
     // Settings detail
     settings_label_sound: 'Suono notifica',
     settings_desc_sound: 'Riproduci un suono quando l\'assistente termina una risposta.',
@@ -3100,6 +3108,10 @@ const LOCALES = {
     transcript: 'トランスクリプト',
     download_transcript: 'Markdown としてダウンロード',
     import: 'インポート',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'セッション全体をJSONとしてエクスポート',
+    import_session_json_tooltip: 'JSONからセッションをインポート',
+    clear_conversation_btn_tooltip: 'この会話のすべてのメッセージをクリア',
     // Settings detail
     settings_label_sound: '通知音',
     settings_desc_sound: 'アシスタントが応答を完了したときに音を鳴らします。',
@@ -4072,6 +4084,10 @@ const LOCALES = {
     transcript: 'Транскрипт',
     download_transcript: 'Скачать как Markdown',
     import: 'Импорт',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Экспортировать сессию как JSON',
+    import_session_json_tooltip: 'Импортировать сессию из JSON',
+    clear_conversation_btn_tooltip: 'Очистить все сообщения в этой беседе',
     settings_label_sound: 'Звук уведомления',
     settings_desc_sound: 'Проигрывать звук, когда помощник завершает ответ.',
     settings_label_notifications: 'Уведомления браузера',
@@ -5193,6 +5209,10 @@ const LOCALES = {
     transcript: 'Transcripción',
     download_transcript: 'Descargar como Markdown',
     import: 'Importar',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Exportar sesión completa como JSON',
+    import_session_json_tooltip: 'Importar sesión desde JSON',
+    clear_conversation_btn_tooltip: 'Borrar todos los mensajes de esta conversación',
     // Settings detail
     settings_label_sound: 'Sonido de notificación',
     settings_desc_sound: 'Reproduce un sonido cuando el asistente termina una respuesta.',
@@ -6287,6 +6307,10 @@ const LOCALES = {
     transcript: 'Protokoll',
     download_transcript: 'Als Markdown herunterladen',
     import: 'Importieren',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Gesamte Sitzung als JSON exportieren',
+    import_session_json_tooltip: 'Sitzung aus JSON importieren',
+    clear_conversation_btn_tooltip: 'Alle Nachrichten in dieser Konversation löschen',
     // Settings detail
     settings_label_sound: 'Benachrichtigungston',
     settings_desc_sound: 'Spielt einen Ton ab, wenn der Assistent eine Antwort beendet.',
@@ -6581,9 +6605,9 @@ const LOCALES = {
     session_toolsets_cleared: 'Toolsets cleared — using global config', // TODO: translate
     session_toolsets_failed: 'Failed to update toolsets: ', // TODO: translate
     session_time_unknown: 'Unbekannt',
-    session_time_minutes_ago: 'Vor {n} Minuten',
-    session_time_hours_ago: 'Vor {n} Stunden',
-    session_time_days_ago: 'Vor {n} Tagen',
+    session_time_minutes_ago: (n) => `Vor ${n} Minuten`,
+    session_time_hours_ago: (n) => `Vor ${n} Stunden`,
+    session_time_days_ago: (n) => `Vor ${n} Tagen`,
     session_time_last_week: 'Letzte Woche',
     session_time_bucket_today: 'Heute',
     session_time_bucket_yesterday: 'Gestern',
@@ -7434,6 +7458,10 @@ const LOCALES = {
     transcript: '记录',
     download_transcript: '下载为 Markdown',
     import: '导入',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: '将会话完整导出为 JSON',
+    import_session_json_tooltip: '从 JSON 导入会话',
+    clear_conversation_btn_tooltip: '清空此会话中的所有消息',
     editing: '编辑中',
     empty_title: '有什么可以帮您？',
     empty_subtitle: '随时提问、运行命令、浏览文件或管理定时任务。',
@@ -8545,6 +8573,10 @@ const LOCALES = {
     transcript: '\u8a18\u9304',
     download_transcript: '\u4e0b\u8f09\u8a18\u9304',
     import: '\u5c0e\u5165',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: '\u5c07\u6703\u8a71\u5b8c\u6574\u532f\u51fa\u70ba JSON',
+    import_session_json_tooltip: '\u5f9e JSON \u532f\u5165\u6703\u8a71',
+    clear_conversation_btn_tooltip: '\u6e05\u7a7a\u6b64\u6703\u8a71\u4e2d\u7684\u6240\u6709\u8a0a\u606f',
     editing: '\u7de8\u8f2f\u4e2d',
     empty_title: '有什麼可以幫忙？',
     empty_subtitle: '點擊上方按鈕開始對話',
@@ -9834,6 +9866,10 @@ const LOCALES = {
     transcript: 'Transcrição',
     download_transcript: 'Baixar como Markdown',
     import: 'Importar',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Exportar sessão completa como JSON',
+    import_session_json_tooltip: 'Importar sessão de JSON',
+    clear_conversation_btn_tooltip: 'Limpar todas as mensagens nesta conversa',
     // Settings detail
     settings_label_sound: 'Som de notificação',
     settings_desc_sound: 'Tocar som quando assistente finalizar resposta.',
@@ -10918,6 +10954,10 @@ const LOCALES = {
     transcript: '대화 기록',
     download_transcript: 'Download as Markdown',
     import: '가져오기',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: '전체 세션을 JSON으로 내보내기',
+    import_session_json_tooltip: 'JSON에서 세션 가져오기',
+    clear_conversation_btn_tooltip: '이 대화의 모든 메시지 지우기',
     // Settings detail
     settings_label_sound: '알림음',
     settings_desc_sound: 'Assistant 응답이 끝나면 소리를 재생합니다.',
@@ -12017,6 +12057,10 @@ const LOCALES = {
     transcript: 'Transcription',
     download_transcript: 'Télécharger en Markdown',
     import: 'Importer',
+    export_session_json: 'JSON',
+    export_session_json_tooltip: 'Exporter la session complète en JSON',
+    import_session_json_tooltip: 'Importer une session depuis JSON',
+    clear_conversation_btn_tooltip: 'Effacer tous les messages de cette conversation',
     settings_label_sound: 'Son de notification',
     settings_desc_sound: 'Jouez un son lorsque l\'assistant termine une réponse.',
     tts_listen: 'Écouter',

static/index.html

@@ -816,9 +816,9 @@
             </div>
             <div class="hermes-action-grid">
               <button class="settings-action-btn" id="btnDownload" title="Download as Markdown" data-i18n-title="download_transcript"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4"/><polyline points="7 10 12 15 17 10"/><line x1="12" y1="15" x2="12" y2="3"/></svg> <span data-i18n="transcript">Transcript</span></button>
-              <button class="settings-action-btn" id="btnExportJSON" title="Export full session as JSON"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M8 3H7a2 2 0 0 0-2 2v5a2 2 0 0 1-2 2 2 2 0 0 1 2 2v5c0 1.1.9 2 2 2h1"/><path d="M16 3h1a2 2 0 0 1 2 2v5a2 2 0 0 0 2 2 2 2 0 0 0-2 2v5a2 2 0 0 1-2 2h-1"/></svg> JSON</button>
-              <button class="settings-action-btn" id="btnImportJSON" title="Import session from JSON"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4"/><polyline points="17 8 12 3 7 8"/><line x1="12" y1="3" x2="12" y2="15"/></svg> <span data-i18n="import">Import</span></button>
-              <button class="settings-action-btn danger" id="btnClearConvModal" onclick="clearConversation()" title="Clear all messages in this conversation"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M3 6h18"/><path d="M19 6v14c0 1-1 2-2 2H7c-1 0-2-1-2-2V6"/><path d="M8 6V4c0-1 1-2 2-2h4c1 0 1 2 2 2v2"/></svg> Clear</button>
+              <button class="settings-action-btn" id="btnExportJSON" title="Export full session as JSON" data-i18n-title="export_session_json_tooltip"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M8 3H7a2 2 0 0 0-2 2v5a2 2 0 0 1-2 2 2 2 0 0 1 2 2v5c0 1.1.9 2 2 2h1"/><path d="M16 3h1a2 2 0 0 1 2 2v5a2 2 0 0 0 2 2 2 2 0 0 0-2 2v5a2 2 0 0 1-2 2h-1"/></svg> <span data-i18n="export_session_json">JSON</span></button>
+              <button class="settings-action-btn" id="btnImportJSON" title="Import session from JSON" data-i18n-title="import_session_json_tooltip"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4"/><polyline points="17 8 12 3 7 8"/><line x1="12" y1="3" x2="12" y2="15"/></svg> <span data-i18n="import">Import</span></button>
+              <button class="settings-action-btn danger" id="btnClearConvModal" onclick="clearConversation()" title="Clear all messages in this conversation" data-i18n-title="clear_conversation_btn_tooltip"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M3 6h18"/><path d="M19 6v14c0 1-1 2-2 2H7c-1 0-2-1-2-2V6"/><path d="M8 6V4c0-1 1-2 2-2h4c1 0 1 2 2 2v2"/></svg> <span data-i18n="clear">Clear</span></button>
             </div>
             <input type="file" id="importFileInput" accept=".json" style="display:none">
           </div>

static/messages.js

@@ -1454,7 +1454,7 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
           const _prevCost=(S.session&&S.session.estimated_cost)||0;
           S.session=d.session;S.messages=d.session.messages||[];if(typeof _messagesTruncated!=='undefined')_messagesTruncated=!!d.session._messages_truncated;
           if(S.session&&S.session.session_id){
-            localStorage.setItem('hermes-webui-session',S.session.session_id);
+            try{localStorage.setItem('hermes-webui-session',S.session.session_id);}catch(_){}
             if(typeof _setActiveSessionUrl==='function') _setActiveSessionUrl(S.session.session_id);
           }
           const _markerOnlyAssistantError=_replaceMarkerOnlyAssistantWithStreamError(S.messages);
@@ -1824,7 +1824,7 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
         clearLiveToolCards();if(!assistantText)removeThinking();
         S.session=session;S.messages=(session.messages||[]).filter(m=>m&&m.role);
         if(S.session&&S.session.session_id){
-          localStorage.setItem('hermes-webui-session',S.session.session_id);
+          try{localStorage.setItem('hermes-webui-session',S.session.session_id);}catch(_){}
           if(typeof _setActiveSessionUrl==='function') _setActiveSessionUrl(S.session.session_id);
         }
         const _markerOnlyAssistantError=_replaceMarkerOnlyAssistantWithStreamError(S.messages);

static/sessions.js

@@ -170,6 +170,14 @@ function _clearSessionCompletionUnread(sid) {
   _saveSessionCompletionUnread();
 }
 
+function _clearSessionViewedCount(sid) {
+  if (!sid) return;
+  const counts = _getSessionViewedCounts();
+  if (!Object.prototype.hasOwnProperty.call(counts, sid)) return;
+  delete counts[sid];
+  _saveSessionViewedCounts();
+}
+
 function _hasSessionCompletionUnread(sid) {
   if (!sid) return false;
   return Object.prototype.hasOwnProperty.call(_getSessionCompletionUnread(), sid);
@@ -416,7 +424,7 @@ async function newSession(flash, options={}){
   S.session=data.session;S.messages=data.session.messages||[];
   S.lastUsage={...(data.session.last_usage||{})};
   if(flash)S.session._flash=true;
-  localStorage.setItem('hermes-webui-session',S.session.session_id);
+  try{localStorage.setItem('hermes-webui-session',S.session.session_id);}catch(_){}
   _setActiveSessionUrl(S.session.session_id);
   _setSessionViewedCount(S.session.session_id, S.session.message_count || 0);
   // Sync chat-header dropdown to the session's model so the UI reflects
@@ -526,7 +534,7 @@ async function loadSession(sid){
   if(typeof syncTopbar==='function') syncTopbar();
   _setSessionViewedCount(S.session.session_id, Number(data.session.message_count || 0));
   _clearSessionCompletionUnread(S.session.session_id);
-  localStorage.setItem('hermes-webui-session',S.session.session_id);
+  try{localStorage.setItem('hermes-webui-session',S.session.session_id);}catch(_){}
   _setActiveSessionUrl(S.session.session_id);
 
   const activeStreamId=S.session.active_stream_id||null;
@@ -810,6 +818,12 @@ function _clearHandoffStorageForSession(sid) {
     _setHandoffStorageValue(sid, _HANDOFF_SUFFIX_DISMISSED_AT, null);
     _setHandoffStorageValue(sid, _HANDOFF_SUFFIX_SUMMARY_HANDLED_AT, null);
   } catch {}
+  // Session deletion should also prune per-session tracking maps. Otherwise
+  // heavy users accumulate one localStorage entry per deleted session forever,
+  // which increases quota pressure and can make future UI persistence fail.
+  try { _clearSessionViewedCount(sid); } catch {}
+  try { _clearSessionCompletionUnread(sid); } catch {}
+  try { _forgetObservedStreamingSession(sid); } catch {}
 }
 
 function _getHandoffDismissedAt(sid) {

static/sw.js

@@ -39,28 +39,35 @@ const SHELL_ASSETS = [
   './manifest.json',
 ];
 
-// Install: pre-cache the app shell
+function deleteOldShellCaches() {
+  return caches.keys().then((keys) =>
+    Promise.all(
+      keys.filter((k) => k !== CACHE_NAME).map((k) => caches.delete(k))
+    )
+  );
+}
+
+// Install: prune old shell caches first, then pre-cache the app shell. Doing
+// this before caches.open(CACHE_NAME) avoids a temporary double-cache window on
+// quota-sensitive browsers during frequent version bumps.
 self.addEventListener('install', (event) => {
   event.waitUntil(
-    caches.open(CACHE_NAME).then((cache) => {
-      return cache.addAll(SHELL_ASSETS).catch((err) => {
-        // Non-fatal: if any asset fails, still activate
-        console.warn('[sw] Shell pre-cache partial failure:', err);
-      });
-    })
+    deleteOldShellCaches().then(() =>
+      caches.open(CACHE_NAME).then((cache) => {
+        return cache.addAll(SHELL_ASSETS).catch((err) => {
+          // Non-fatal: if any asset fails, still activate
+          console.warn('[sw] Shell pre-cache partial failure:', err);
+        });
+      })
+    )
   );
   self.skipWaiting();
 });
 
-// Activate: clean up old caches
+// Activate: keep the old-cache cleanup as a safety net in case install was
+// interrupted or an older worker was already waiting.
 self.addEventListener('activate', (event) => {
-  event.waitUntil(
-    caches.keys().then((keys) =>
-      Promise.all(
-        keys.filter((k) => k !== CACHE_NAME).map((k) => caches.delete(k))
-      )
-    )
-  );
+  event.waitUntil(deleteOldShellCaches());
   self.clients.claim();
 });
 

static/ui.js

@@ -2614,7 +2614,10 @@ function renderMd(raw){
     const parseHeader=r=>r.trim().replace(/^\|/,'').replace(/\|$/,'').split('|').map(c=>`<th>${inlineMd(c.trim())}</th>`).join('');
     const header=`<tr>${parseHeader(rows[0])}</tr>`;
     const body=rows.slice(2).map(r=>`<tr>${parseRow(r)}</tr>`).join('');
-    return `<table><thead>${header}</thead><tbody>${body}</tbody></table>`;
+    // Surround with blank lines so the final paragraph splitter treats the
+    // generated table as its own block even when the regex consumes one of the
+    // markdown block's trailing newlines.
+    return `\n\n<table><thead>${header}</thead><tbody>${body}</tbody></table>\n\n`;
   });
   // #487: Outer image pass — handles ![alt](url) in plain paragraphs (outside tables/lists).
   // Runs AFTER the table pass (images in table cells are handled by inlineMd() above).
@@ -2757,7 +2760,7 @@ function renderMd(raw){
     return '\x00E'+(_pre_stash.length-1)+'\x00';
   });
   const parts=s.split(/\n{2,}/);
-  s=parts.map(p=>{p=p.trim();if(!p)return '';if(/^<(h[1-6]|ul|ol|pre|hr|blockquote)|^\x00[EQ]/.test(p))return p;return `<p>${p.replace(/\n/g,'<br>')}</p>`;}).join('\n');
+  s=parts.map(p=>{p=p.trim();if(!p)return '';if(/^<(h[1-6]|ul|ol|table|pre|hr|blockquote)|^\x00[EQ]/.test(p))return p;return `<p>${p.replace(/\n/g,'<br>')}</p>`;}).join('\n');
   s=s.replace(/\x00E(\d+)\x00/g,(_,i)=>_pre_stash[+i]);
   // ── Restore MEDIA stash → inline images or download links ─────────────────
   s=s.replace(/\x00D(\d+)\x00/g,(_,i)=>{

tests/conftest.py

@@ -171,6 +171,36 @@ def pytest_configure(config):
 # imports trigger botocore initialisation.
 os.environ.setdefault("AWS_EC2_METADATA_DISABLED", "true")
 
+# ── Permanent os.execv guard for the pytest session ────────────────────────
+# Several tests in tests/test_update_banner_fixes.py exercise
+# api.updates._schedule_restart(), which spawns a DAEMON thread that sleeps
+# for a short delay and then calls ``os.execv(sys.executable, sys.argv)``.
+# Those tests monkeypatch ``os.execv`` to a no-op for the test scope, but
+# monkeypatch teardown happens at test exit — if the daemon thread has not
+# yet woken up by then (system load, GC pause, _apply_lock contention), the
+# real ``os.execv`` is restored before the thread fires it. The daemon then
+# REPLACES the pytest process image with a fresh ``pytest tests/ -q ...``
+# invocation, looking from the outside like pytest "hangs at 99%" and then
+# restarts the entire suite from 0% — a self-perpetuating loop.
+#
+# Daemon threads cannot be reliably joined from a test fixture (they live in
+# ``api.updates`` module scope), so the only safe answer is to render
+# ``os.execv`` permanently inert for the pytest session. Production code is
+# unaffected because production never imports this conftest.
+#
+# Tests that need to verify execv WAS called still monkeypatch it themselves
+# — their patched version takes precedence over this no-op wrapper for the
+# test's lifetime, and the no-op only kicks in after teardown for daemon
+# threads that wake up late.
+_real_execv = os.execv
+
+def _pytest_session_safe_execv(_exe, _args):  # pragma: no cover — never called in prod
+    # Drop the call on the floor. A late-firing daemon thread from
+    # _schedule_restart() must not be able to re-exec the pytest process.
+    return None
+
+os.execv = _pytest_session_safe_execv
+
 # ── Hermetic network isolation ─────────────────────────────────────────────
 # Tests must not reach the public internet. Outbound to Anthropic / OpenAI /
 # Amazon / OpenRouter / etc. is forbidden by default. The test suite already

tests/test_issue2386_localstorage_quota.py

@@ -0,0 +1,34 @@
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def _script(path):
+    return (ROOT / path).read_text()
+
+
+def _assert_storage_setitem_guarded(src, needle):
+    matches = [line.strip() for line in src.splitlines() if needle in line]
+    assert matches, f"expected at least one {needle} write"
+    for line in matches:
+        assert line.startswith("try{localStorage.setItem("), (
+            f"localStorage quota errors must not escape from {needle} writes: {line}"
+        )
+        assert "catch(_)" in line or "catch(e)" in line or "catch{}" in line
+
+
+def test_active_session_localstorage_writes_ignore_quota_errors():
+    """Session persistence writes are best-effort when the browser quota is full (#2386)."""
+    for path in ["static/sessions.js", "static/commands.js", "static/messages.js"]:
+        _assert_storage_setitem_guarded(
+            _script(path),
+            "localStorage.setItem('hermes-webui-session'",
+        )
+
+
+def test_workspace_panel_localstorage_write_ignores_quota_errors():
+    """Workspace panel state should not break UI toggles if localStorage throws (#2386)."""
+    _assert_storage_setitem_guarded(
+        _script("static/boot.js"),
+        "localStorage.setItem('hermes-webui-workspace-panel'",
+    )

tests/test_issue2389_storage_pressure.py

@@ -0,0 +1,59 @@
+"""Regression coverage for storage-pressure cleanup from issue #2389."""
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+SW_SRC = (ROOT / "static" / "sw.js").read_text(encoding="utf-8")
+SESSIONS_SRC = (ROOT / "static" / "sessions.js").read_text(encoding="utf-8")
+
+
+def _function_block(src: str, name: str, window: int = 1600) -> str:
+    idx = src.find(f"function {name}(")
+    assert idx != -1, f"missing function {name}"
+    return src[idx : idx + window]
+
+
+def test_service_worker_install_deletes_old_caches_before_opening_new_cache():
+    install_idx = SW_SRC.find("self.addEventListener('install'")
+    assert install_idx != -1, "service worker must define an install handler"
+    install_block = SW_SRC[install_idx : SW_SRC.find("self.addEventListener('activate'", install_idx)]
+    cleanup_idx = install_block.find("deleteOldShellCaches().then")
+    open_idx = install_block.find("caches.open(CACHE_NAME)")
+    assert cleanup_idx != -1, "install must delete stale shell caches before pre-cache"
+    assert open_idx != -1, "install must still pre-cache the current shell cache"
+    assert cleanup_idx < open_idx, (
+        "opening the new shell cache before deleting old ones creates a temporary "
+        "double-cache window that increases quota pressure"
+    )
+
+
+def test_service_worker_keeps_activate_cleanup_safety_net():
+    activate_idx = SW_SRC.find("self.addEventListener('activate'")
+    assert activate_idx != -1, "service worker must define an activate handler"
+    activate_block = SW_SRC[activate_idx : activate_idx + 500]
+    assert "event.waitUntil(deleteOldShellCaches())" in activate_block
+    assert "self.clients.claim()" in activate_block
+
+
+def test_deleted_sessions_prune_all_session_tracking_maps():
+    assert "const SESSION_VIEWED_COUNTS_KEY = 'hermes-session-viewed-counts';" in SESSIONS_SRC
+    assert "const SESSION_COMPLETION_UNREAD_KEY = 'hermes-session-completion-unread';" in SESSIONS_SRC
+    assert "const SESSION_OBSERVED_STREAMING_KEY = 'hermes-session-observed-streaming';" in SESSIONS_SRC
+    assert "function _clearSessionViewedCount(sid)" in SESSIONS_SRC
+
+    clear_block = _function_block(SESSIONS_SRC, "_clearHandoffStorageForSession")
+    assert "_clearSessionViewedCount(sid)" in clear_block
+    assert "_clearSessionCompletionUnread(sid)" in clear_block
+    assert "_forgetObservedStreamingSession(sid)" in clear_block
+
+
+def test_session_viewed_count_prune_is_best_effort_and_persists_when_changed():
+    viewed_block = _function_block(SESSIONS_SRC, "_clearSessionViewedCount")
+    assert "Object.prototype.hasOwnProperty.call(counts, sid)" in viewed_block
+    assert "delete counts[sid]" in viewed_block
+    assert "_saveSessionViewedCounts()" in viewed_block
+
+    clear_block = _function_block(SESSIONS_SRC, "_clearHandoffStorageForSession")
+    assert "try { _clearSessionViewedCount(sid); } catch {}" in clear_block
+    assert "try { _clearSessionCompletionUnread(sid); } catch {}" in clear_block
+    assert "try { _forgetObservedStreamingSession(sid); } catch {}" in clear_block

tests/test_native_image_attachments.py

@@ -17,6 +17,7 @@ from api.streaming import (
     _attachment_name,
     _build_native_multimodal_message,
     _NATIVE_IMAGE_MAX_BYTES,
+    _sanitize_messages_for_api,
 )
 from api.routes import _normalize_chat_attachments
 
@@ -318,6 +319,41 @@ class TestBuildNativeMultimodalMessage:
             assert data_url.startswith('data:image/png;base64,')
             assert len(result) == 2
 
+    def test_text_image_mode_strips_historical_image_url_parts(self):
+        """#2297: text-only providers must not replay old native image parts."""
+        history = [
+            {
+                'role': 'user',
+                'content': [
+                    {'type': 'text', 'text': 'what is in this image?'},
+                    {'type': 'image_url', 'image_url': {'url': 'data:image/png;base64,AAA='}},
+                ],
+                'attachments': [{'name': 'photo.png'}],
+                'timestamp': 123,
+            },
+            {'role': 'assistant', 'content': 'It is a chart.'},
+        ]
+        cfg = {'agent': {'image_input_mode': 'text'}}
+
+        sanitized = _sanitize_messages_for_api(history, cfg=cfg)
+
+        assert sanitized[0] == {'role': 'user', 'content': 'what is in this image?'}
+        assert 'image_url' not in str(sanitized)
+        assert 'attachments' not in sanitized[0]
+        assert sanitized[1] == {'role': 'assistant', 'content': 'It is a chart.'}
+
+    def test_native_image_mode_keeps_historical_image_url_parts(self):
+        """Vision-capable/native mode keeps existing multimodal history intact."""
+        content = [
+            {'type': 'text', 'text': 'describe'},
+            {'type': 'image_url', 'image_url': {'url': 'data:image/png;base64,AAA='}},
+        ]
+        cfg = {'agent': {'image_input_mode': 'native'}}
+
+        sanitized = _sanitize_messages_for_api([{'role': 'user', 'content': content}], cfg=cfg)
+
+        assert sanitized == [{'role': 'user', 'content': content}]
+
     def test_fake_png_rejected_by_magic_bytes(self):
         """A file named .png that is not actually an image must be rejected."""
         with TemporaryDirectory() as d:

tests/test_pytest_execv_guard.py

@@ -0,0 +1,35 @@
+"""Regression guard for the pytest "hangs at 99% then restarts from 0%" loop.
+
+Root cause documented in tests/conftest.py — daemon threads spawned by
+api.updates._schedule_restart() can fire os.execv() AFTER monkeypatch
+teardown restores the real os.execv, which re-execs the entire pytest
+process. The conftest installs a permanent no-op wrapper on os.execv that
+shadows any late-firing daemon thread.
+
+This test pins the guard so a future conftest refactor can't silently
+remove it.
+"""
+import os
+
+
+def test_conftest_installs_permanent_execv_guard():
+    """os.execv must be replaced by the conftest's safe no-op wrapper."""
+    # The wrapper is named `_pytest_session_safe_execv` in conftest.py.
+    # Verify the module attribute now points to that wrapper, not the real
+    # libc-bound function.
+    assert os.execv.__name__ == '_pytest_session_safe_execv', (
+        f"os.execv must be the conftest-installed pytest-safe no-op, but "
+        f"resolves to {os.execv!r}. Did a recent conftest refactor remove "
+        f"the guard? See conftest.py § 'Permanent os.execv guard for the "
+        f"pytest session' — without it, late-firing _schedule_restart "
+        f"daemon threads re-exec pytest and the suite loops forever."
+    )
+
+
+def test_safe_execv_returns_none_does_not_exec():
+    """The wrapper must be a true no-op — it must not raise, exec, or block."""
+    # Pass deliberately bogus args to confirm the wrapper drops them rather
+    # than passing them through to the real execv.
+    result = os.execv('/nonexistent/binary/path/that/should/not/be/executed',
+                      ['/nonexistent/binary/path/that/should/not/be/executed'])
+    assert result is None

tests/test_renderer_js_behaviour.py

@@ -187,6 +187,37 @@ class TestRendererSanitization:
 
 class TestCommonLLMShapes:
 
+    def test_commonmark_table_is_not_wrapped_in_paragraph(self, driver_path):
+        src = (
+            "| 升级时段 | 人数 |\n"
+            "|---------|------|\n"
+            "| 5/15（发布当天） | ~30 人 |\n"
+            "| 5/16（今天） | ~10 人 |"
+        )
+        out = _render(driver_path, src)
+        assert "<table><thead>" in out
+        assert "<th>升级时段</th>" in out
+        assert "<td>5/15（发布当天）</td>" in out
+        assert "<td>~10 人</td>" in out
+        assert "<p><table" not in out, (
+            f"Markdown tables are block elements and must not be paragraph-wrapped: {out!r}"
+        )
+
+    def test_table_between_paragraphs_stays_block_level(self, driver_path):
+        src = (
+            "Before the table.\n\n"
+            "| Key | Value |\n"
+            "| --- | --- |\n"
+            "| A | B |\n\n"
+            "After the table."
+        )
+        out = _render(driver_path, src)
+        assert "<p>Before the table.</p>" in out
+        assert "<table><thead>" in out
+        assert "<p>After the table.</p>" in out
+        assert "<p><table" not in out
+        assert "</table></p>" not in out
+
     def test_strikethrough_outside_quote(self, driver_path):
         out = _render(driver_path, "This was ~~outdated~~ but is now fine.")
         assert "<del>outdated</del>" in out

tests/test_run_journal_frontend_static.py

@@ -35,3 +35,60 @@ def test_frontend_replay_cursor_uses_eventsource_last_event_id():
     assert "source.addEventListener(_runJournalEventName,_rememberRunJournalCursor)" in MESSAGES_SRC
     assert "after_seq=${encodeURIComponent(String(_runJournalReplayAfterSeq()))}" in MESSAGES_SRC
     assert "after_seq=0" not in MESSAGES_SRC
+
+
+def test_replayed_long_task_events_enter_the_same_live_timeline_handlers():
+    """Run-journal replay must not grow a parallel long-task renderer.
+
+    The run-state consistency contract depends on replayed journal events
+    flowing through the same EventSource handlers as live streams.  Otherwise a
+    live long task can render as Thinking -> progress text -> tool cards, while
+    the same journaled event sequence replays as a flattened or reordered scene.
+    """
+    wire_pos = MESSAGES_SRC.index("function _wireSSE(source)")
+    wire_block = MESSAGES_SRC[wire_pos : MESSAGES_SRC.index("async function _restoreSettledSession", wire_pos)]
+    replay_events = [
+        "reasoning",
+        "interim_assistant",
+        "tool",
+        "tool_complete",
+        "compressing",
+        "compressed",
+        "metering",
+        "done",
+        "apperror",
+    ]
+
+    for event_name in replay_events:
+        assert f"source.addEventListener('{event_name}'" in wire_block, (
+            f"{event_name} must be handled by the shared live/replay SSE pipeline"
+        )
+
+    assert "updateThinking(" in wire_block, "reasoning replay should use the live Thinking card path"
+    assert "appendLiveToolCard(tc)" in wire_block, "tool replay should use live tool-card rendering"
+    assert "setCompressionUi({" in wire_block, "compression replay should use the compression card path"
+    assert "_runJournalReplayParams()" in MESSAGES_SRC, "replay attachments should enter _wireSSE via EventSource"
+
+
+def test_run_journal_cursor_tracks_every_long_task_timeline_event():
+    """Every user-visible long-task event needs cursor tracking for parity replay."""
+    cursor_loop_pos = MESSAGES_SRC.index("for(const _runJournalEventName of [")
+    cursor_loop = MESSAGES_SRC[cursor_loop_pos : MESSAGES_SRC.index("]", cursor_loop_pos)]
+    timeline_events = [
+        "token",
+        "interim_assistant",
+        "reasoning",
+        "tool",
+        "tool_complete",
+        "compressing",
+        "compressed",
+        "metering",
+        "done",
+        "apperror",
+        "cancel",
+    ]
+
+    for event_name in timeline_events:
+        assert f"'{event_name}'" in cursor_loop, (
+            f"{event_name} must advance the replay cursor to avoid duplicate timeline replay"
+        )

tests/test_session_index.py

@@ -369,6 +369,44 @@ def test_pre_compression_snapshot_hidden_from_active_sidebar_but_file_remains(mo
     assert [row["session_id"] for row in rows] == ["new_sid"]
 
 
+def test_fuller_pre_compression_snapshot_replaces_shorter_visible_segment(monkeypatch):
+    """If the hidden snapshot has the fuller transcript, keep it reachable.
+
+    Auto-compression can leave a visible continuation segment in the sidebar
+    while the fuller transcript remains on disk marked as a pre-compression
+    snapshot. In that case the default session list should prefer the fuller
+    transcript so the conversation does not look like recent messages vanished.
+    """
+    snapshot = Session(
+        session_id="full_parent",
+        title="Long Conversation",
+        messages=[
+            {"role": "user", "content": "first"},
+            {"role": "assistant", "content": "second"},
+            {"role": "user", "content": "latest user"},
+            {"role": "assistant", "content": "latest answer"},
+        ],
+        pre_compression_snapshot=True,
+        updated_at=300.0,
+    )
+    continuation = Session(
+        session_id="short_child",
+        title="Long Conversation",
+        messages=[{"role": "user", "content": "first"}],
+        parent_session_id="full_parent",
+        updated_at=400.0,
+    )
+    snapshot.save()
+    continuation.save()
+    monkeypatch.setattr(models, "_enrich_sidebar_lineage_metadata", lambda _sessions: None)
+
+    rows = models.all_sessions()
+
+    assert [row["session_id"] for row in rows] == ["full_parent"]
+    assert rows[0]["message_count"] == 4
+    assert rows[0]["pre_compression_snapshot"] is True
+
+
 def test_session_save_does_not_persist_metadata_message_count_hint():
     s = Session(
         session_id="sess_private_hint",

tests/test_session_sidebar_relative_time.py

@@ -169,3 +169,12 @@ def test_relative_time_strings_are_localized_in_english_and_spanish_bundles():
         "session_time_bucket_older",
     ):
         assert key in I18N_JS
+
+
+def test_german_relative_time_translations_interpolate_numbers():
+    assert "session_time_minutes_ago: (n) => `Vor ${n} Minuten`" in I18N_JS
+    assert "session_time_hours_ago: (n) => `Vor ${n} Stunden`" in I18N_JS
+    assert "session_time_days_ago: (n) => `Vor ${n} Tagen`" in I18N_JS
+    assert "session_time_minutes_ago: 'Vor {n} Minuten'" not in I18N_JS
+    assert "session_time_hours_ago: 'Vor {n} Stunden'" not in I18N_JS
+    assert "session_time_days_ago: 'Vor {n} Tagen'" not in I18N_JS

tests/test_session_sidecar_repair.py

@@ -231,7 +231,7 @@ class TestRepairStalePendingNoDeadlock:
 class TestDraftRecovery:
     """When no core transcript exists, the pending user message is restored as
     a recovered user turn (_recovered=True) and the error marker says
-    'Previous turn did not complete.' — NOT 'preserved as a draft'."""
+    a clear restart interruption marker — NOT 'preserved as a draft'."""
 
     def test_pending_message_recovered_as_user_turn(self, hermes_home, monkeypatch):
         """When core transcript is missing, the pending_user_message is appended
@@ -310,7 +310,10 @@ class TestDraftRecovery:
         assert "preserved as a draft" not in content, (
             f"Error marker should not say 'preserved as a draft', got: {content}"
         )
-        assert "Previous turn did not complete" in content
+        assert "Response interrupted" in content
+        assert "WebUI process restarted" in content
+        assert "user message above was preserved" in content
+        assert error_msgs[0].get("type") == "interrupted"
 
     def test_pending_attachments_recovered(self, hermes_home, monkeypatch):
         """Attachments on the pending message are carried over to the recovered turn."""
@@ -604,7 +607,9 @@ class TestNonEmptyMessagesPendingCleared:
         # Exactly one error marker
         error_msgs = [m for m in s.messages if m.get("_error")]
         assert len(error_msgs) == 1
-        assert "Previous turn did not complete" in error_msgs[0]["content"]
+        assert "Response interrupted" in error_msgs[0]["content"]
+        assert "WebUI process restarted" in error_msgs[0]["content"]
+        assert error_msgs[0].get("type") == "interrupted"
 
         # Pending fields fully cleared
         assert s.pending_user_message is None

tests/test_sprint2.py

@@ -67,7 +67,7 @@ def test_raw_endpoint_path_traversal_blocked(cleanup_test_sessions):
         get_raw(f"/api/file/raw?session_id={sid}&path=../../etc/passwd")
         assert False
     except urllib.error.HTTPError as e:
-        assert e.code in (400, 500)
+        assert e.code in (400, 404, 500)
 
 def test_raw_endpoint_missing_file_returns_404(cleanup_test_sessions):
     sid, _ = make_session_tracked(cleanup_test_sessions)

tests/test_sprint46.py

@@ -415,10 +415,15 @@ def test_manual_compress_worker_uses_session_profile_env(monkeypatch, tmp_path,
         seen_env = None
 
         def __init__(self, **kwargs):
+            from api.config import _thread_ctx
+
             skill_module = sys.modules.get("tools.skills_tool")
+            thread_env = getattr(_thread_ctx, "env", {})
             EnvAssertingAgent.seen_env = {
                 "HERMES_HOME": os.environ.get("HERMES_HOME"),
                 "HERMES_TEST_PROFILE_ENV": os.environ.get("HERMES_TEST_PROFILE_ENV"),
+                "THREAD_HERMES_HOME": thread_env.get("HERMES_HOME"),
+                "THREAD_HERMES_TEST_PROFILE_ENV": thread_env.get("HERMES_TEST_PROFILE_ENV"),
                 "SKILL_MODULE_HOME": getattr(skill_module, "HERMES_HOME", None),
                 "SKILL_MODULE_DIR": getattr(skill_module, "SKILLS_DIR", None),
             }
@@ -461,6 +466,8 @@ def test_manual_compress_worker_uses_session_profile_env(monkeypatch, tmp_path,
     assert EnvAssertingAgent.seen_env == {
         "HERMES_HOME": str(profile_home),
         "HERMES_TEST_PROFILE_ENV": "work-runtime",
+        "THREAD_HERMES_HOME": str(profile_home),
+        "THREAD_HERMES_TEST_PROFILE_ENV": "work-runtime",
         "SKILL_MODULE_HOME": profile_home,
         "SKILL_MODULE_DIR": profile_home / "skills",
     }

tests/test_sprint6.py

@@ -1,5 +1,5 @@
 """Sprint 6 tests: Escape from editor, Phase D validation, HTML extraction, cron create, session export."""
-import json, uuid, pathlib, urllib.request, urllib.error
+import json, uuid, pathlib, urllib.parse, urllib.request, urllib.error
 REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
 
 from tests._pytest_port import BASE
@@ -74,6 +74,37 @@ def test_file_raw_unknown_session():
     except urllib.error.HTTPError as e:
         assert e.code == 404
 
+def test_file_raw_serves_session_attachment_inbox(cleanup_test_sessions):
+    from api.upload import _session_attachment_dir
+
+    sid, workspace = make_session_tracked(cleanup_test_sessions)
+    filename = f"uploaded-chat-image-{uuid.uuid4().hex}.png"
+    attachment_dir = _session_attachment_dir(sid)
+    attachment_dir.mkdir(parents=True, exist_ok=True)
+    payload = b"fake-png-bytes"
+    (attachment_dir / filename).write_bytes(payload)
+
+    assert not (workspace / filename).exists(), "regression must exercise attachment fallback"
+    raw, headers, status = get_raw(
+        f"/api/file/raw?session_id={sid}&path={urllib.parse.quote(filename)}"
+    )
+    assert status == 200
+    assert raw == payload
+    assert "image/png" in headers.get("Content-Type", "")
+
+def test_file_raw_attachment_fallback_rejects_traversal(cleanup_test_sessions):
+    from api.upload import _session_attachment_dir
+
+    sid, _ = make_session_tracked(cleanup_test_sessions)
+    attachment_dir = _session_attachment_dir(sid)
+    attachment_dir.mkdir(parents=True, exist_ok=True)
+    (attachment_dir / "safe.txt").write_text("safe", encoding="utf-8")
+    try:
+        get_raw(f"/api/file/raw?session_id={sid}&path={urllib.parse.quote('../../safe.txt')}")
+        assert False, "Expected 404"
+    except urllib.error.HTTPError as e:
+        assert e.code == 404
+
 # ── Cron create ──
 
 def test_cron_create_requires_prompt():

tests/test_title_aux_routing.py

@@ -502,6 +502,64 @@ class TestBackgroundTitleProfileRouting(unittest.TestCase):
         self.assertEqual(getattr(fake_skill_module, 'SKILLS_DIR'), 'default-home/skills')
         self.assertEqual(mock_session.title, 'Profile Routed Title')
 
+    def test_background_profile_env_routes_load_config_and_provider_credentials(self):
+        """Hybrid worker env must satisfy config and os.getenv provider-key readers."""
+        import tempfile
+
+        import pytest
+
+        import api.profiles as profiles
+        from api.config import _thread_ctx
+        try:
+            from hermes_cli import config as hermes_config
+        except ModuleNotFoundError:
+            pytest.skip('hermes_cli is not installed in this CI environment')
+
+        session = types.SimpleNamespace(profile='work')
+        captured = {}
+
+        with tempfile.TemporaryDirectory() as tmp:
+            default_home = os.path.join(tmp, 'default-home')
+            profile_home = os.path.join(tmp, 'profile-home')
+            os.makedirs(default_home, exist_ok=True)
+            os.makedirs(profile_home, exist_ok=True)
+            with open(os.path.join(default_home, 'config.yaml'), 'w', encoding='utf-8') as f:
+                f.write('model:\n  provider: default-provider\n  default: default-model\n')
+            with open(os.path.join(profile_home, 'config.yaml'), 'w', encoding='utf-8') as f:
+                f.write('model:\n  provider: profile-provider\n  default: profile-model\n')
+
+            with patch('api.profiles.get_hermes_home_for_profile', return_value=profile_home):
+                runtime_env = {
+                    'PROFILE_ONLY_KEY': 'profile-only',
+                    'OPENROUTER_API_KEY': 'profile-openrouter-key',
+                }
+                with patch('api.profiles.get_profile_runtime_env', return_value=runtime_env):
+                    with patch.dict(os.environ, {'HERMES_HOME': default_home, 'OPENROUTER_API_KEY': 'default-openrouter-key'}, clear=False):
+                        os.environ.pop('PROFILE_ONLY_KEY', None)
+                        hermes_config._LOAD_CONFIG_CACHE.clear()
+                        with profiles.profile_env_for_background_worker(session, 'background title'):
+                            loaded = hermes_config.load_config()
+                            captured['loaded_provider'] = loaded.get('model', {}).get('provider')
+                            captured['process_home'] = os.environ.get('HERMES_HOME')
+                            captured['process_runtime_key'] = os.environ.get('PROFILE_ONLY_KEY')
+                            captured['provider_credential'] = os.getenv('OPENROUTER_API_KEY')
+                            captured['thread_home'] = getattr(_thread_ctx, 'env', {}).get('HERMES_HOME')
+                            captured['thread_runtime_key'] = getattr(_thread_ctx, 'env', {}).get('PROFILE_ONLY_KEY')
+                        captured['restored_home'] = os.environ.get('HERMES_HOME')
+                        captured['restored_runtime_key'] = os.environ.get('PROFILE_ONLY_KEY')
+                        captured['restored_provider_credential'] = os.environ.get('OPENROUTER_API_KEY')
+                        hermes_config._LOAD_CONFIG_CACHE.clear()
+
+        self.assertEqual(captured['loaded_provider'], 'profile-provider')
+        self.assertEqual(captured['process_home'], profile_home)
+        self.assertEqual(captured['process_runtime_key'], 'profile-only')
+        self.assertEqual(captured['provider_credential'], 'profile-openrouter-key')
+        self.assertEqual(captured['thread_home'], profile_home)
+        self.assertEqual(captured['thread_runtime_key'], 'profile-only')
+        self.assertEqual(captured['restored_home'], default_home)
+        self.assertIsNone(captured['restored_runtime_key'])
+        self.assertEqual(captured['restored_provider_credential'], 'default-openrouter-key')
+
 
 class TestAuxTitleTimeoutEdgeCases(unittest.TestCase):
     """_aux_title_timeout must reject zero, negative, and non-numeric values."""

tests/test_update_banner_fixes.py

@@ -476,9 +476,12 @@ class TestUpdateSummaryRouteModelSelection:
         monkeypatch.setattr(cfg, 'get_effective_default_model', lambda: 'openai/test-main')
 
         def fake_resolve_model_provider(model):
+            thread_env = getattr(cfg._thread_ctx, 'env', {})
             captured['model_resolution_env'] = {
                 'HERMES_HOME': os.environ.get('HERMES_HOME'),
                 'HERMES_TEST_PROFILE_ENV': os.environ.get('HERMES_TEST_PROFILE_ENV'),
+                'THREAD_HERMES_HOME': thread_env.get('HERMES_HOME'),
+                'THREAD_HERMES_TEST_PROFILE_ENV': thread_env.get('HERMES_TEST_PROFILE_ENV'),
             }
             return model, 'openai', 'https://example.test/v1'
 
@@ -514,9 +517,12 @@ class TestUpdateSummaryRouteModelSelection:
                         )
 
         def fake_get_text_auxiliary_client(task, main_runtime=None):
+            thread_env = getattr(cfg._thread_ctx, 'env', {})
             captured['aux_env'] = {
                 'HERMES_HOME': os.environ.get('HERMES_HOME'),
                 'HERMES_TEST_PROFILE_ENV': os.environ.get('HERMES_TEST_PROFILE_ENV'),
+                'THREAD_HERMES_HOME': thread_env.get('HERMES_HOME'),
+                'THREAD_HERMES_TEST_PROFILE_ENV': thread_env.get('HERMES_TEST_PROFILE_ENV'),
                 'SKILL_MODULE_HOME': getattr(fake_skill_module, 'HERMES_HOME'),
                 'SKILL_MODULE_DIR': getattr(fake_skill_module, 'SKILLS_DIR'),
             }
@@ -564,10 +570,14 @@ class TestUpdateSummaryRouteModelSelection:
         assert captured['model_resolution_env'] == {
             'HERMES_HOME': str(profile_home),
             'HERMES_TEST_PROFILE_ENV': 'work-runtime',
+            'THREAD_HERMES_HOME': str(profile_home),
+            'THREAD_HERMES_TEST_PROFILE_ENV': 'work-runtime',
         }
         assert captured['aux_env'] == {
             'HERMES_HOME': str(profile_home),
             'HERMES_TEST_PROFILE_ENV': 'work-runtime',
+            'THREAD_HERMES_HOME': str(profile_home),
+            'THREAD_HERMES_TEST_PROFILE_ENV': 'work-runtime',
             'SKILL_MODULE_HOME': profile_home,
             'SKILL_MODULE_DIR': profile_home / 'skills',
         }