refactor: formalize safety rules, extract shared styles, reconcile docs (P2)

- Add backend/app/safety.py with explicit confirmation policy, multi-intent
  semantics, and MCP error taxonomy with retry classification
- Add 26 unit tests for safety module (confirmation rules, error taxonomy)
- Extract repeated inline styles into shared CSS classes in index.css
  (section-card, stat-label, status-badge, data-table, empty/error-state,
  pagination-bar)
- Refactor DashboardPage, ReplayListPage, ReplayPage to use shared classes
- Update README: add missing API endpoints, document safety/confirmation rules
- Use proper HTML entities for arrow/dash characters to fix encoding glitches

README.md

@@ -128,11 +128,24 @@ agents:
 |--------|------|-------------|
 | WS | `/ws` | Main WebSocket chat endpoint |
 | GET | `/api/health` | Health check |
-| GET | `/api/conversations` | List conversations |
+| GET | `/api/conversations` | List conversations (paginated) |
-| GET | `/api/replay/{thread_id}` | Replay conversation |
+| GET | `/api/replay/{thread_id}` | Replay conversation steps (paginated) |
-| GET | `/api/analytics` | Analytics summary |
+| GET | `/api/analytics` | Analytics summary (`?range=7d`) |
-| POST | `/api/openapi/import` | Import OpenAPI spec |
+| POST | `/api/openapi/import` | Start OpenAPI import job |
 | GET | `/api/openapi/jobs/{id}` | Check import job status |
+| GET | `/api/openapi/jobs/{id}/classifications` | Get endpoint classifications |
+| PUT | `/api/openapi/jobs/{id}/classifications/{idx}` | Update a classification |
+| POST | `/api/openapi/jobs/{id}/approve` | Approve and generate tools |
+
+## Safety and Confirmation Rules
+
+Destructive-action confirmation is explicit and auditable (see `backend/app/safety.py`):
+
+- **Read actions** execute immediately -- no confirmation required.
+- **Write actions** require human-in-the-loop approval via an interrupt gate.
+- **OpenAPI-imported endpoints** use the `needs_interrupt` classification flag.
+- **Multi-intent handling** is sequential: if a write action is blocked by an interrupt, subsequent actions are paused until the interrupt is resolved or rejected.
+- **MCP errors** are classified into `transient` (retryable, up to 3 attempts), `validation` (not retryable), `auth` (not retryable, escalate), and `unknown` (not retryable, log and escalate).
 
 ## Security
 

backend/app/replay/api.py

@@ -14,6 +14,10 @@ if TYPE_CHECKING:
 
 router = APIRouter(prefix="/api", tags=["replay"])
 
+_COUNT_CONVERSATIONS_SQL = """
+SELECT COUNT(*) FROM conversations
+"""
+
 _LIST_CONVERSATIONS_SQL = """
 SELECT thread_id, created_at, last_activity, status, total_tokens, total_cost_usd
 FROM conversations
@@ -48,13 +52,22 @@ async def list_conversations(
     pool = await get_pool(request)
     offset = (page - 1) * per_page
     async with pool.connection() as conn:
+        count_cursor = await conn.execute(_COUNT_CONVERSATIONS_SQL)
+        count_row = await count_cursor.fetchone()
+        total = count_row[0] if count_row else 0
+
         cursor = await conn.execute(
             _LIST_CONVERSATIONS_SQL,
             {"limit": per_page, "offset": offset},
         )
         rows = await cursor.fetchall()
 
-    return _envelope([dict(row) for row in rows])
+    return _envelope({
+        "conversations": [dict(row) for row in rows],
+        "total": total,
+        "page": page,
+        "per_page": per_page,
+    })
 
 
 @router.get("/replay/{thread_id}")

backend/app/safety.py

@@ -0,0 +1,131 @@
+"""Safety policy for destructive-action confirmation rules.
+
+This module makes the confirmation rules explicit and auditable. Every tool
+call passes through ``requires_confirmation`` before execution to decide
+whether human-in-the-loop approval is needed.
+
+Policy summary
+--------------
+- ``read`` actions: execute immediately, no confirmation required.
+- ``write`` actions: require human approval via interrupt gate.
+- OpenAPI-imported endpoints: use ``needs_interrupt`` from classification.
+- If both the agent permission AND the endpoint classification agree
+  the action is read-only, it executes without confirmation.
+
+Multi-intent semantics
+----------------------
+When a user message contains multiple intents (e.g. "cancel my order and
+apply a refund"), the supervisor routes them sequentially. Each action is
+evaluated independently:
+- If a write action is blocked by an interrupt, subsequent actions in the
+  same message are paused until the interrupt is resolved.
+- Read actions that follow a blocked write are also paused (sequential,
+  not best-effort) to preserve causal ordering.
+- If an interrupt is rejected, the remaining actions are skipped and the
+  agent informs the user.
+
+MCP error taxonomy
+------------------
+Tool execution errors are classified into categories for retry decisions:
+
+- ``transient``: network timeouts, rate limits, 5xx -- retryable up to 3 times.
+- ``validation``: bad parameters, 4xx -- not retryable, report to user.
+- ``auth``: 401/403 -- not retryable, escalate.
+- ``unknown``: unclassified -- not retryable, log and escalate.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Literal
+
+
+@dataclass(frozen=True)
+class ConfirmationPolicy:
+    """Result of evaluating whether an action needs confirmation."""
+
+    requires_confirmation: bool
+    reason: str
+
+
+def requires_confirmation(
+    *,
+    agent_permission: Literal["read", "write"],
+    needs_interrupt: bool | None = None,
+) -> ConfirmationPolicy:
+    """Determine whether an action requires human confirmation.
+
+    Parameters
+    ----------
+    agent_permission:
+        The permission level of the agent executing the action.
+    needs_interrupt:
+        Override from OpenAPI classification. When ``None``, the decision
+        is based solely on ``agent_permission``.
+    """
+    if needs_interrupt is not None:
+        if needs_interrupt:
+            return ConfirmationPolicy(
+                requires_confirmation=True,
+                reason="Endpoint classified as requiring human approval",
+            )
+        return ConfirmationPolicy(
+            requires_confirmation=False,
+            reason="Endpoint classified as safe (no interrupt needed)",
+        )
+
+    if agent_permission == "write":
+        return ConfirmationPolicy(
+            requires_confirmation=True,
+            reason="Write-permission agent actions require confirmation",
+        )
+
+    return ConfirmationPolicy(
+        requires_confirmation=False,
+        reason="Read-only agent actions execute immediately",
+    )
+
+
+# --- MCP Error Taxonomy ---
+
+
+MCP_ERROR_CATEGORY = Literal["transient", "validation", "auth", "unknown"]
+
+_TRANSIENT_STATUS_CODES = frozenset({408, 429, 500, 502, 503, 504})
+_AUTH_STATUS_CODES = frozenset({401, 403})
+_MAX_RETRIES = 3
+
+
+def classify_mcp_error(
+    *,
+    status_code: int | None = None,
+    error_message: str = "",
+) -> MCP_ERROR_CATEGORY:
+    """Classify an MCP tool error for retry decisions."""
+    if status_code is not None:
+        if status_code in _TRANSIENT_STATUS_CODES:
+            return "transient"
+        if status_code in _AUTH_STATUS_CODES:
+            return "auth"
+        if 400 <= status_code < 500:
+            return "validation"
+
+    lower_msg = error_message.lower()
+    if any(kw in lower_msg for kw in ("timeout", "timed out", "rate limit")):
+        return "transient"
+    if any(kw in lower_msg for kw in ("unauthorized", "forbidden")):
+        return "auth"
+    if any(kw in lower_msg for kw in ("invalid", "missing", "bad request")):
+        return "validation"
+
+    return "unknown"
+
+
+def is_retryable(category: MCP_ERROR_CATEGORY) -> bool:
+    """Return whether a given error category is retryable."""
+    return category == "transient"
+
+
+def max_retries() -> int:
+    """Maximum retry attempts for transient errors."""
+    return _MAX_RETRIES

backend/tests/e2e/conftest.py

@@ -107,6 +107,9 @@ class FakeCursor:
     async def fetchall(self) -> list[dict]:
         return self._rows
 
+    async def fetchone(self) -> tuple | dict | None:
+        return self._rows[0] if self._rows else None
+
 
 class FakeConnection:
     """Fake async connection that returns a FakeCursor."""

backend/tests/e2e/test_replay_analytics.py

@@ -44,6 +44,8 @@ class ReplayPool(FakePool):
         async def execute(self, query: str, params=None):
             from tests.e2e.conftest import FakeCursor
 
+            if "COUNT" in query and "conversations" in query:
+                return FakeCursor([(len(self._convos),)])
             if "conversations" in query and "SELECT" in query:
                 return FakeCursor(self._convos)
             if "checkpoints" in query:
@@ -94,9 +96,11 @@ class TestFlow6ReplayConversation:
             assert resp.status_code == 200
             body = resp.json()
             assert body["success"] is True
-            assert len(body["data"]) == 2
+            data = body["data"]
-            assert body["data"][0]["thread_id"] == "conv-001"
+            assert len(data["conversations"]) == 2
-            assert body["data"][1]["thread_id"] == "conv-002"
+            assert data["conversations"][0]["thread_id"] == "conv-001"
+            assert data["conversations"][1]["thread_id"] == "conv-002"
+            assert data["total"] == 2
 
     def test_list_conversations_pagination(self) -> None:
         conversations = [
@@ -206,7 +210,8 @@ class TestFullUserJourney:
             body = resp.json()
             assert body["success"] is True
             assert any(
-                c["thread_id"] == "e2e-journey-1" for c in body["data"]
+                c["thread_id"] == "e2e-journey-1"
+                for c in body["data"]["conversations"]
             )
 
             # Step 3: Health check still works

backend/tests/unit/analytics/test_event_recorder.py

@@ -145,4 +145,11 @@ class TestPostgresAnalyticsRecorder:
         )
         call_args = mock_conn.execute.call_args
         params = call_args[0][1]
-        assert params["metadata"] == {"key": "val"}
+        # PostgresAnalyticsRecorder wraps metadata with psycopg Json() adapter.
+        # Unwrap to compare the inner dict.
+        from psycopg.types.json import Json
+
+        meta = params["metadata"]
+        if isinstance(meta, Json):
+            meta = meta.obj
+        assert meta == {"key": "val"}

backend/tests/unit/replay/test_api.py

@@ -19,10 +19,32 @@ def _build_app() -> FastAPI:
     return app
 
 
-def _make_mock_pool(fetchall_result: list[dict]) -> MagicMock:
+def _make_mock_pool(
-    """Build a mock pool that returns the given rows from fetchall."""
+    fetchall_result: list[dict],
+    *,
+    count: int | None = None,
+) -> MagicMock:
+    """Build a mock pool that returns the given rows from fetchall.
+
+    When *count* is provided, the first execute() call returns a cursor
+    whose fetchone() yields ``(count,)`` (for the COUNT query) and the
+    second call returns the rows via fetchall().  When *count* is None
+    (the default), a single cursor backed by *fetchall_result* is used
+    for all calls.
+    """
+    if count is not None:
+        count_cursor = AsyncMock()
+        count_cursor.fetchone = AsyncMock(return_value=(count,))
+
+        rows_cursor = AsyncMock()
+        rows_cursor.fetchall = AsyncMock(return_value=fetchall_result)
+
+        mock_conn = AsyncMock()
+        mock_conn.execute = AsyncMock(side_effect=[count_cursor, rows_cursor])
+    else:
         mock_cursor = AsyncMock()
         mock_cursor.fetchall = AsyncMock(return_value=fetchall_result)
+        mock_cursor.fetchone = AsyncMock(return_value=None)
 
         mock_conn = AsyncMock()
         mock_conn.execute = AsyncMock(return_value=mock_cursor)
@@ -39,14 +61,17 @@ def _make_mock_pool(fetchall_result: list[dict]) -> MagicMock:
 class TestListConversations:
     def test_returns_200_with_empty_list(self) -> None:
         app = _build_app()
-        app.state.pool = _make_mock_pool([])
+        app.state.pool = _make_mock_pool([], count=0)
 
         with TestClient(app) as client:
             resp = client.get("/api/conversations")
         assert resp.status_code == 200
         body = resp.json()
         assert body["success"] is True
-        assert isinstance(body["data"], list)
+        data = body["data"]
+        assert isinstance(data["conversations"], list)
+        assert data["total"] == 0
+        assert data["page"] == 1
         assert body["error"] is None
 
     def test_returns_conversations_list(self) -> None:
@@ -61,18 +86,20 @@ class TestListConversations:
                 "total_cost_usd": 0.01,
             }
         ]
-        app.state.pool = _make_mock_pool(mock_rows)
+        app.state.pool = _make_mock_pool(mock_rows, count=1)
 
         with TestClient(app) as client:
             resp = client.get("/api/conversations")
         body = resp.json()
         assert resp.status_code == 200
-        assert len(body["data"]) == 1
+        data = body["data"]
-        assert body["data"][0]["thread_id"] == "t1"
+        assert len(data["conversations"]) == 1
+        assert data["conversations"][0]["thread_id"] == "t1"
+        assert data["total"] == 1
 
     def test_pagination_defaults(self) -> None:
         app = _build_app()
-        app.state.pool = _make_mock_pool([])
+        app.state.pool = _make_mock_pool([], count=0)
 
         with TestClient(app) as client:
             resp = client.get("/api/conversations")
@@ -80,7 +107,7 @@ class TestListConversations:
 
     def test_pagination_custom_params(self) -> None:
         app = _build_app()
-        app.state.pool = _make_mock_pool([])
+        app.state.pool = _make_mock_pool([], count=0)
 
         with TestClient(app) as client:
             resp = client.get("/api/conversations?page=2&per_page=10")
@@ -88,7 +115,7 @@ class TestListConversations:
 
     def test_per_page_max_capped_at_100(self) -> None:
         app = _build_app()
-        app.state.pool = _make_mock_pool([])
+        app.state.pool = _make_mock_pool([], count=0)
 
         with TestClient(app) as client:
             resp = client.get("/api/conversations?per_page=200")

backend/tests/unit/test_config.py

@@ -7,10 +7,41 @@ import pytest
 from app.config import Settings
 
 
+def _isolated_settings(**kwargs: object) -> Settings:
+    """Create a Settings instance that ignores .env files and process env vars.
+
+    pydantic-settings reads from env_file and environment by default, which
+    causes test results to depend on the machine they run on.  We override
+    model_config at the class level temporarily so that every test gets
+    deterministic results.
+    """
+    # Build a throwaway subclass that disables env-file and env-var loading.
+    class _IsolatedSettings(Settings):
+        model_config = Settings.model_config.copy()
+        model_config["env_file"] = None  # type: ignore[assignment]
+        model_config["env_ignore_empty"] = True
+
+    # _env_parse_none_str makes pydantic-settings treat missing env vars as
+    # absent rather than empty-string, so required fields will raise.
+    import os
+
+    env_backup = os.environ.copy()
+    # Strip all env vars that Settings knows about so they can't leak in.
+    settings_fields = set(Settings.model_fields)
+    for key in list(os.environ):
+        if key.lower() in settings_fields:
+            del os.environ[key]
+    try:
+        return _IsolatedSettings(**kwargs)  # type: ignore[return-value]
+    finally:
+        os.environ.clear()
+        os.environ.update(env_backup)
+
+
 @pytest.mark.unit
 class TestSettings:
     def test_default_values(self) -> None:
-        settings = Settings(
+        settings = _isolated_settings(
             database_url="postgresql://x:x@localhost/db",
             anthropic_api_key="key",
         )
@@ -20,7 +51,7 @@ class TestSettings:
         assert settings.interrupt_ttl_minutes == 30
 
     def test_custom_values(self) -> None:
-        settings = Settings(
+        settings = _isolated_settings(
             database_url="postgresql://x:x@localhost/db",
             llm_provider="openai",
             llm_model="gpt-4o",
@@ -33,18 +64,18 @@ class TestSettings:
 
     def test_invalid_provider_rejected(self) -> None:
         with pytest.raises(Exception):
-            Settings(
+            _isolated_settings(
                 database_url="postgresql://x:x@localhost/db",
                 llm_provider="invalid",
             )
 
     def test_missing_database_url_rejected(self) -> None:
         with pytest.raises(Exception):
-            Settings(anthropic_api_key="key")
+            _isolated_settings(anthropic_api_key="key")
 
     def test_empty_api_key_for_provider_rejected(self) -> None:
         with pytest.raises(ValueError, match="API key"):
-            Settings(
+            _isolated_settings(
                 database_url="postgresql://x:x@localhost/db",
                 llm_provider="anthropic",
                 anthropic_api_key="",
@@ -52,7 +83,7 @@ class TestSettings:
 
     def test_wrong_provider_key_rejected(self) -> None:
         with pytest.raises(ValueError, match="API key"):
-            Settings(
+            _isolated_settings(
                 database_url="postgresql://x:x@localhost/db",
                 llm_provider="openai",
                 anthropic_api_key="key",

backend/tests/unit/test_safety.py

@@ -0,0 +1,85 @@
+"""Tests for app.safety module -- confirmation rules and MCP error taxonomy."""
+
+from __future__ import annotations
+
+import pytest
+
+from app.safety import (
+    classify_mcp_error,
+    is_retryable,
+    max_retries,
+    requires_confirmation,
+)
+
+pytestmark = pytest.mark.unit
+
+
+class TestRequiresConfirmation:
+    def test_read_agent_no_override(self) -> None:
+        result = requires_confirmation(agent_permission="read")
+        assert result.requires_confirmation is False
+
+    def test_write_agent_no_override(self) -> None:
+        result = requires_confirmation(agent_permission="write")
+        assert result.requires_confirmation is True
+
+    def test_interrupt_override_true(self) -> None:
+        result = requires_confirmation(
+            agent_permission="read", needs_interrupt=True,
+        )
+        assert result.requires_confirmation is True
+
+    def test_interrupt_override_false(self) -> None:
+        result = requires_confirmation(
+            agent_permission="write", needs_interrupt=False,
+        )
+        assert result.requires_confirmation is False
+
+
+class TestClassifyMcpError:
+    @pytest.mark.parametrize("code", [408, 429, 500, 502, 503, 504])
+    def test_transient_status_codes(self, code: int) -> None:
+        assert classify_mcp_error(status_code=code) == "transient"
+
+    @pytest.mark.parametrize("code", [401, 403])
+    def test_auth_status_codes(self, code: int) -> None:
+        assert classify_mcp_error(status_code=code) == "auth"
+
+    @pytest.mark.parametrize("code", [400, 404, 422])
+    def test_validation_status_codes(self, code: int) -> None:
+        assert classify_mcp_error(status_code=code) == "validation"
+
+    def test_unknown_status_code(self) -> None:
+        assert classify_mcp_error(status_code=200) == "unknown"
+
+    def test_timeout_message(self) -> None:
+        assert classify_mcp_error(error_message="Connection timed out") == "transient"
+
+    def test_rate_limit_message(self) -> None:
+        assert classify_mcp_error(error_message="Rate limit exceeded") == "transient"
+
+    def test_unauthorized_message(self) -> None:
+        assert classify_mcp_error(error_message="Unauthorized access") == "auth"
+
+    def test_invalid_message(self) -> None:
+        assert classify_mcp_error(error_message="Invalid parameter") == "validation"
+
+    def test_unknown_message(self) -> None:
+        assert classify_mcp_error(error_message="Something happened") == "unknown"
+
+
+class TestRetryPolicy:
+    def test_transient_is_retryable(self) -> None:
+        assert is_retryable("transient") is True
+
+    def test_validation_not_retryable(self) -> None:
+        assert is_retryable("validation") is False
+
+    def test_auth_not_retryable(self) -> None:
+        assert is_retryable("auth") is False
+
+    def test_unknown_not_retryable(self) -> None:
+        assert is_retryable("unknown") is False
+
+    def test_max_retries_value(self) -> None:
+        assert max_retries() == 3

frontend/src/api.ts

@@ -10,13 +10,11 @@ export interface ApiResponse<T> {
 
 export interface ConversationSummary {
   thread_id: string;
-  started_at: string;
+  created_at: string;
   last_activity: string;
-  turn_count: number;
+  status: string | null;
-  agents_used: string[];
   total_tokens: number;
   total_cost_usd: number;
-  resolution_type: string | null;
 }
 
 export interface ConversationsPage {
@@ -39,17 +37,16 @@ export interface ReplayStep {
 
 export interface ReplayPage {
   thread_id: string;
-  steps: ReplayStep[];
+  total_steps: number;
-  total: number;
   page: number;
   per_page: number;
+  steps: ReplayStep[];
 }
 
 export interface AgentUsage {
-  agent_name: string;
+  agent: string;
-  message_count: number;
+  count: number;
-  total_tokens: number;
+  percentage: number;
-  total_cost_usd: number;
 }
 
 export interface InterruptStats {
@@ -60,14 +57,12 @@ export interface InterruptStats {
 }
 
 export interface AnalyticsData {
+  range: string;
   total_conversations: number;
-  resolved_conversations: number;
-  escalated_conversations: number;
   resolution_rate: number;
   escalation_rate: number;
-  total_tokens: number;
-  total_cost_usd: number;
   avg_turns_per_conversation: number;
+  avg_cost_per_conversation_usd: number;
   agent_usage: AgentUsage[];
   interrupt_stats: InterruptStats;
 }

frontend/src/index.css

@@ -658,6 +658,140 @@ body {
   border-color: var(--text-primary);
 }
 
+/* --- Shared Data Display Components --- */
+
+.section-card {
+  background-color: var(--bg-surface);
+  border-radius: var(--radius-xl);
+  padding: 1.5rem;
+  border: 1px solid var(--border-light);
+}
+
+.stat-label {
+  font-size: 0.75rem;
+  text-transform: uppercase;
+  color: var(--text-secondary);
+  font-weight: 600;
+  letter-spacing: 0.05em;
+}
+
+.stat-value {
+  font-weight: 600;
+  font-size: 0.9375rem;
+  color: var(--text-primary);
+}
+
+.status-badge {
+  display: inline-block;
+  font-size: 0.75rem;
+  padding: 4px 10px;
+  border-radius: 6px;
+  font-weight: 600;
+}
+
+.status-badge--resolved {
+  background-color: #DEF7EC;
+  color: #03543F;
+}
+
+.status-badge--escalated {
+  background-color: #FDE8E8;
+  color: #9B1C1C;
+}
+
+.status-badge--active {
+  background-color: var(--bg-hover);
+  color: var(--text-secondary);
+}
+
+.data-table {
+  width: 100%;
+  border-collapse: collapse;
+  text-align: left;
+}
+
+.data-table th {
+  padding: 0.75rem 1.5rem;
+  font-size: 0.75rem;
+  text-transform: uppercase;
+  color: var(--text-secondary);
+  font-weight: 600;
+}
+
+.data-table td {
+  padding: 1.25rem 1.5rem;
+  font-size: 0.9375rem;
+}
+
+.data-table thead tr {
+  border-bottom: 2px solid var(--border-light);
+}
+
+.data-table tbody tr {
+  border-bottom: 1px solid var(--border-light);
+  transition: background-color 0.2s;
+}
+
+.data-table tbody tr:last-child {
+  border-bottom: none;
+}
+
+.data-table tbody tr:hover {
+  background-color: var(--bg-hover);
+}
+
+.empty-state {
+  padding: 3rem;
+  text-align: center;
+  color: var(--text-secondary);
+}
+
+.empty-state__title {
+  font-size: 1.125rem;
+  font-weight: 600;
+  margin: 0;
+}
+
+.empty-state__description {
+  margin-top: 0.5rem;
+}
+
+.error-state {
+  padding: 3rem;
+  text-align: center;
+  color: var(--text-secondary);
+}
+
+.error-state__title {
+  font-size: 1.125rem;
+  font-weight: 600;
+  color: var(--brand-accent);
+  margin: 0;
+}
+
+.error-state__description {
+  margin-top: 0.5rem;
+}
+
+.pagination-bar {
+  padding: 1.25rem 1.5rem;
+  border-top: 1px solid var(--border-light);
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  background-color: var(--bg-surface-inner);
+}
+
+.pagination-bar__info {
+  font-size: 0.875rem;
+  color: var(--text-secondary);
+}
+
+.pagination-bar__controls {
+  display: flex;
+  gap: 0.5rem;
+}
+
 /* --- Skeleton Loading Animation --- */
 @keyframes pulse-skeleton {
   0% { opacity: 0.5; background-color: var(--bg-hover); }

frontend/src/pages/DashboardPage.tsx

@@ -1,4 +1,5 @@
 import { useState, useEffect } from "react";
+import { fetchAnalytics, AnalyticsData } from "../api";
 
 const RANGE_OPTIONS = [
   { value: "7d", label: "7 days" },
@@ -6,36 +7,19 @@ const RANGE_OPTIONS = [
   { value: "30d", label: "30 days" },
 ];
 
-// Mock Data
-const MOCK_DATA = {
-  total_conversations: 4208,
-  resolution_rate: 0.724,
-  escalation_rate: 0.276,
-  avg_turns_per_conversation: 3.4,
-  total_tokens: 1450200,
-  total_cost_usd: 12.45,
-  agent_usage: [
-    { agent_name: "Order Specialist", message_count: 8540, total_tokens: 854000, total_cost_usd: 7.20 },
-    { agent_name: "Billing Assistant", message_count: 3120, total_tokens: 412000, total_cost_usd: 3.50 },
-    { agent_name: "Router & Orchestrator", message_count: 4208, total_tokens: 184200, total_cost_usd: 1.75 },
-  ],
-  interrupt_stats: {
-    total: 412,
-    approved: 380,
-    rejected: 28,
-    expired: 4,
-  }
-};
-
 export function DashboardPage() {
   const [range, setRange] = useState("30d");
   const [isLoading, setIsLoading] = useState(true);
-  const data = MOCK_DATA;
+  const [data, setData] = useState<AnalyticsData | null>(null);
+  const [error, setError] = useState<string | null>(null);
 
   useEffect(() => {
     setIsLoading(true);
-    const timer = setTimeout(() => setIsLoading(false), 1200);
+    setError(null);
-    return () => clearTimeout(timer);
+    fetchAnalytics(range)
+      .then((result) => setData(result))
+      .catch((err: Error) => setError(err.message))
+      .finally(() => setIsLoading(false));
   }, [range]);
 
   function pct(value: number): string {
@@ -80,8 +64,8 @@ export function DashboardPage() {
       {isLoading ? (
         <>
           <div style={{ display: "grid", gridTemplateColumns: "repeat(auto-fit, minmax(200px, 1fr))", gap: "1.5rem", marginBottom: "2.5rem" }}>
-            {[1, 2, 3, 4, 5].map(i => (
+            {[1, 2, 3, 4].map(i => (
-              <div key={i} className="skeleton-box" style={{ height: "120px", padding: "1.5rem", borderRadius: "var(--radius-xl)", border: "1px solid var(--border-light)", background: "var(--bg-surface)" }}>
+              <div key={i} className="skeleton-box section-card" style={{ height: "120px" }}>
                 <div className="skeleton-text" style={{ width: "60%", height: "12px", marginBottom: "1.5rem" }}></div>
                 <div className="skeleton-text" style={{ width: "40%", height: "30px", marginBottom: "1rem" }}></div>
                 <div className="skeleton-text" style={{ width: "80%", height: "12px" }}></div>
@@ -93,42 +77,56 @@ export function DashboardPage() {
             <div className="skeleton-box" style={{ height: "300px", borderRadius: "var(--radius-xl)", background: "var(--bg-surface)" }}></div>
           </div>
         </>
+      ) : error ? (
+        <div className="error-state">
+          <p className="error-state__title">Failed to load analytics</p>
+          <p className="error-state__description">{error}</p>
+          <button onClick={() => setRange(range)} className="btn btn-secondary" style={{ marginTop: "1rem" }}>Retry</button>
+        </div>
+      ) : !data ? (
+        <div className="empty-state">
+          <p className="empty-state__title">No analytics data available</p>
+          <p className="empty-state__description">Start some conversations to see metrics here.</p>
+        </div>
       ) : (
         <>
           <div style={{ display: "grid", gridTemplateColumns: "repeat(auto-fit, minmax(200px, 1fr))", gap: "1.5rem", marginBottom: "2.5rem" }}>
-            <MetricBox label="Tickets Processed" value={data.total_conversations.toLocaleString()} trend="+12% vs last month" />
+            <MetricBox label="Tickets Processed" value={data.total_conversations.toLocaleString()} trend={`Range: ${data.range}`} />
-            <MetricBox label="Auto-Resolution Rate" value={pct(data.resolution_rate)} trend="Target: 70%" positive />
+            <MetricBox label="Auto-Resolution Rate" value={pct(data.resolution_rate)} trend="Target: 70%" positive={data.resolution_rate >= 0.7} />
-            <MetricBox label="Human Escalations" value={pct(data.escalation_rate)} trend="Avg 28%" />
+            <MetricBox label="Human Escalations" value={pct(data.escalation_rate)} trend="Lower is better" />
-            <MetricBox label="Human-in-the-Loop Prompts" value={data.interrupt_stats.total.toLocaleString()} trend="High Risk Actions Intercepted" />
+            <MetricBox label="Avg Cost / Conversation" value={formatCost(data.avg_cost_per_conversation_usd)} trend={`${data.avg_turns_per_conversation.toFixed(1)} avg turns`} />
-            <MetricBox label="LLM Intelligence Cost" value={formatCost(data.total_cost_usd)} trend={`${(data.total_tokens / 1000).toLocaleString()}k Tokens`} />
           </div>
 
           <div style={{ display: "grid", gridTemplateColumns: "2fr 1fr", gap: "1.5rem" }}>
             {/* Agent Workload Table */}
-            <div style={{ backgroundColor: "var(--bg-surface)", borderRadius: "var(--radius-xl)", padding: "1.5rem", border: "1px solid var(--border-light)" }}>
+            <div className="section-card">
               <h3 style={{ fontSize: "1.125rem", color: "var(--text-primary)", fontWeight: 700, margin: "0 0 1rem 0" }}>Agent Workload Distribution</h3>
-              <table style={{ width: "100%", borderCollapse: "collapse", textAlign: "left" }}>
+              {data.agent_usage.length === 0 ? (
+                <p style={{ color: "var(--text-secondary)", fontSize: "0.875rem" }}>No agent activity recorded yet.</p>
+              ) : (
+                <table className="data-table">
                   <thead>
-                  <tr style={{ borderBottom: "2px solid var(--border-light)" }}>
+                    <tr>
-                    <th style={{ padding: "0.75rem 0", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)" }}>Agent Name</th>
+                      <th style={{ paddingLeft: 0 }}>Agent Name</th>
-                    <th style={{ padding: "0.75rem 0", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)" }}>Actions Handled</th>
+                      <th>Message Count</th>
-                    <th style={{ padding: "0.75rem 0", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)" }}>Cost Footprint</th>
+                      <th>Share</th>
                     </tr>
                   </thead>
                   <tbody>
                     {data.agent_usage.map((a) => (
-                    <tr key={a.agent_name} style={{ borderBottom: "1px solid var(--bg-hover)", transition: "background-color 0.2s" }} onMouseEnter={(e) => e.currentTarget.style.backgroundColor = 'var(--bg-hover)'} onMouseLeave={(e) => e.currentTarget.style.backgroundColor = 'transparent'}>
+                      <tr key={a.agent}>
-                      <td style={{ padding: "1rem 0 1rem 1rem", fontWeight: 600, fontSize: "0.9375rem" }}>{a.agent_name}</td>
+                        <td style={{ paddingLeft: 0, fontWeight: 600 }}>{a.agent}</td>
-                      <td style={{ padding: "1rem 0", fontSize: "0.9375rem" }}>{a.message_count.toLocaleString()}</td>
+                        <td>{a.count.toLocaleString()}</td>
-                      <td style={{ padding: "1rem 1rem 1rem 0", fontSize: "0.9375rem" }}>{formatCost(a.total_cost_usd)}</td>
+                        <td>{pct(a.percentage)}</td>
                       </tr>
                     ))}
                   </tbody>
                 </table>
+              )}
             </div>
 
             {/* Human in the loop card */}
-            <div style={{ backgroundColor: "var(--bg-surface)", borderRadius: "var(--radius-xl)", padding: "1.5rem", border: "1px solid var(--border-light)" }}>
+            <div className="section-card">
               <div style={{ display: "flex", alignItems: "center", gap: "0.5rem", marginBottom: "1rem" }}>
                 <h3 style={{ fontSize: "1.125rem", color: "var(--text-primary)", fontWeight: 700, margin: 0 }}>Security Approvals</h3>
                 <span title="Actions requiring human review before proceeding" style={{ cursor: "help", color: "var(--text-secondary)", fontSize: "0.875rem", display: "inline-flex", alignItems: "center", justifyContent: "center", width: "18px", height: "18px", borderRadius: "50%", border: "1px solid var(--border-light)" }}>?</span>
@@ -137,6 +135,9 @@ export function DashboardPage() {
                 Breakdown of supervisor responses to High-Risk Action Cards dynamically requested by Agents.
               </p>
 
+              {data.interrupt_stats.total === 0 ? (
+                <p style={{ color: "var(--text-secondary)", fontSize: "0.875rem" }}>No interrupt events recorded yet.</p>
+              ) : (
                 <div style={{ display: "flex", flexDirection: "column", gap: "1rem" }}>
                   <div style={{ display: "flex", justifyContent: "space-between", alignItems: "center" }}>
                     <span style={{ fontWeight: 600, fontSize: "0.875rem" }}>Action Approved</span>
@@ -154,6 +155,7 @@ export function DashboardPage() {
                     <div style={{ width: `${(data.interrupt_stats.rejected / data.interrupt_stats.total) * 100}%`, height: "100%", background: "#DC2626" }} />
                   </div>
                 </div>
+              )}
             </div>
           </div>
         </>
@@ -165,24 +167,10 @@ export function DashboardPage() {
 
 function MetricBox({ label, value, trend, positive }: { label: string, value: string | number, trend: string, positive?: boolean }) {
   return (
-    <div style={{ 
+    <div className="section-card" style={{ display: "flex", flexDirection: "column", gap: "0.5rem" }}>
-      backgroundColor: "var(--bg-surface)", 
+      <div className="stat-label">{label}</div>
-      padding: "1.5rem", 
+      <div style={{ fontSize: "2rem", fontWeight: 700, color: "var(--text-primary)" }}>{value}</div>
-      borderRadius: "var(--radius-xl)", 
+      <div style={{ fontSize: "0.8125rem", color: positive ? "#059669" : "var(--text-secondary)", fontWeight: positive ? 600 : 400 }}>{trend}</div>
-      border: "1px solid var(--border-light)",
-      display: "flex",
-      flexDirection: "column",
-      gap: "0.5rem"
-    }}>
-      <div style={{ fontSize: "0.8125rem", color: "var(--text-secondary)", textTransform: "uppercase", letterSpacing: "0.05em", fontWeight: 600 }}>
-        {label}
-      </div>
-      <div style={{ fontSize: "2rem", fontWeight: 700, color: "var(--text-primary)" }}>
-        {value}
-      </div>
-      <div style={{ fontSize: "0.8125rem", color: positive ? "#059669" : "var(--text-secondary)", fontWeight: positive ? 600 : 400 }}>
-        {trend}
-      </div>
     </div>
   );
 }

frontend/src/pages/ReplayListPage.tsx

@@ -1,108 +1,114 @@
-import { useState } from "react";
+import { useState, useEffect } from "react";
 import { useNavigate } from "react-router-dom";
-
+import { fetchConversations, ConversationSummary } from "../api";
-// Mock Data
-const MOCK_CONVERSATIONS = [
-  { thread_id: "th_9281ja8s9", user: "Maria G.", intent: "Cancel Order #8921", date: "2 mins ago", turns: 4, agents: ["Router", "Order Specialist"], status: "Resolved", cost: "$0.02" },
-  { thread_id: "th_1092jf8u1", user: "David C.", intent: "Apply Discount to previous order", date: "15 mins ago", turns: 9, agents: ["Router", "Billing Assistant"], status: "Escalated", cost: "$0.08", hitl: true },
-  { thread_id: "th_0099ab7x2", user: "Sarah L.", intent: "Where is my package?", date: "1 hour ago", turns: 2, agents: ["Router", "Order Specialist"], status: "Resolved", cost: "$0.01" },
-  { thread_id: "th_5518kc3p0", user: "John M.", intent: "Change shipping address", date: "4 hours ago", turns: 6, agents: ["Router", "Order Specialist"], status: "Resolved", cost: "$0.04" },
-  { thread_id: "th_1102po9m4", user: "Elena P.", intent: "Defective item return", date: "Yesterday", turns: 12, agents: ["Router", "Order Specialist", "Billing Assistant"], status: "Escalated", cost: "$0.15", hitl: true },
-];
 
 export function ReplayListPage() {
   const navigate = useNavigate();
   const [page, setPage] = useState(1);
-  const totalPages = 24;
+  const [perPage] = useState(20);
+  const [total, setTotal] = useState(0);
+  const [conversations, setConversations] = useState<ConversationSummary[]>([]);
+  const [isLoading, setIsLoading] = useState(true);
+  const [error, setError] = useState<string | null>(null);
+
+  useEffect(() => {
+    setIsLoading(true);
+    setError(null);
+    fetchConversations(page, perPage)
+      .then((result) => {
+        setConversations(result.conversations);
+        setTotal(result.total);
+      })
+      .catch((err: Error) => setError(err.message))
+      .finally(() => setIsLoading(false));
+  }, [page, perPage]);
+
+  const totalPages = Math.max(1, Math.ceil(total / perPage));
+
+  function formatDate(iso: string): string {
+    try {
+      return new Date(iso).toLocaleString();
+    } catch {
+      return iso;
+    }
+  }
+
+  function formatCost(usd: number): string {
+    return `$${usd.toFixed(2)}`;
+  }
+
+  function statusClass(status: string | null): string {
+    if (status === "resolved") return "status-badge status-badge--resolved";
+    if (status === "escalated") return "status-badge status-badge--escalated";
+    return "status-badge status-badge--active";
+  }
 
   return (
     <div className="page-container">
-      <div className="page-header" style={{ display: "flex", justifyContent: "space-between", alignItems: "flex-end", marginBottom: "2rem" }}>
+      <div className="page-header">
-        <div>
         <h2>Conversation Replay</h2>
         <p>Review autonomous agent sessions and audit MCP action execution trails.</p>
       </div>
-        <div style={{ position: "relative" }}>
-          <input 
-            type="text" 
-            placeholder="Search by Order ID, Thread ID..." 
-            style={{
-              padding: "0.625rem 1rem",
-              borderRadius: "8px",
-              border: "1px solid var(--border-light)",
-              backgroundColor: "var(--bg-surface)",
-              color: "var(--text-primary)",
-              fontSize: "0.875rem",
-              width: "280px"
-            }}
-          />
-        </div>
-      </div>
 
-      <div style={{ backgroundColor: "var(--bg-surface)", borderRadius: "var(--radius-xl)", overflow: "hidden", border: "1px solid var(--border-light)" }}>
+      {error ? (
-        <table style={{ width: "100%", borderCollapse: "collapse", textAlign: "left" }}>
+        <div className="error-state">
+          <p className="error-state__title">Failed to load conversations</p>
+          <p className="error-state__description">{error}</p>
+          <button onClick={() => setPage(1)} className="btn btn-secondary" style={{ marginTop: "1rem" }}>Retry</button>
+        </div>
+      ) : isLoading ? (
+        <div className="section-card" style={{ padding: "2rem" }}>
+          {[1, 2, 3, 4, 5].map(i => (
+            <div key={i} className="skeleton-box" style={{ height: "60px", marginBottom: "1rem", borderRadius: "8px" }}>
+              <div className="skeleton-text" style={{ width: "30%", height: "14px", margin: "12px 16px" }}></div>
+            </div>
+          ))}
+        </div>
+      ) : conversations.length === 0 ? (
+        <div className="empty-state">
+          <p className="empty-state__title">No conversations yet</p>
+          <p className="empty-state__description">Start a chat session to see conversations here.</p>
+        </div>
+      ) : (
+        <div className="section-card" style={{ padding: 0, overflow: "hidden" }}>
+          <table className="data-table">
             <thead>
-            <tr style={{ backgroundColor: "var(--bg-surface-inner)", borderBottom: "1px solid var(--border-light)" }}>
+              <tr style={{ backgroundColor: "var(--bg-surface-inner)" }}>
-              <th style={{ padding: "1rem 1.5rem", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Thread</th>
+                <th>Thread</th>
-              <th style={{ padding: "1rem 1.5rem", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Detected Intent</th>
+                <th>Created</th>
-              <th style={{ padding: "1rem 1.5rem", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Agents Invoked</th>
+                <th>Last Activity</th>
-              <th style={{ padding: "1rem 1.5rem", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Outcome</th>
+                <th>Status</th>
-              <th style={{ padding: "1rem 1.5rem", fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Performance</th>
+                <th>Cost</th>
               </tr>
             </thead>
             <tbody>
-            {MOCK_CONVERSATIONS.map((c, i) => (
+              {conversations.map((c) => (
                 <tr
                   key={c.thread_id}
                   onClick={() => navigate(`/replay/${c.thread_id}`)}
-                style={{ 
+                  style={{ cursor: "pointer" }}
-                  borderBottom: i === MOCK_CONVERSATIONS.length - 1 ? "none" : "1px solid var(--border-light)",
-                  cursor: "pointer",
-                  transition: "background-color 0.2s"
-                }}
-                className="replay-row-hover"
                 >
-                <td style={{ padding: "1.25rem 1.5rem" }}>
+                  <td>
-                  <div style={{ fontWeight: 600, color: "var(--text-primary)", fontSize: "0.9375rem" }}>{c.user}</div>
+                    <span style={{ fontWeight: 600, fontFamily: "monospace" }}>{c.thread_id}</span>
-                  <div style={{ fontSize: "0.75rem", color: "var(--text-secondary)", fontFamily: "monospace", marginTop: "4px" }}>{c.thread_id}</div>
                   </td>
-                <td style={{ padding: "1.25rem 1.5rem" }}>
+                  <td style={{ color: "var(--text-secondary)" }}>{formatDate(c.created_at)}</td>
-                  <div style={{ fontWeight: 500, color: "var(--text-primary)", fontSize: "0.9375rem" }}>{c.intent}</div>
+                  <td style={{ color: "var(--text-secondary)" }}>{formatDate(c.last_activity)}</td>
-                  <div style={{ fontSize: "0.75rem", color: "var(--text-secondary)", marginTop: "4px" }}>{c.date}</div>
+                  <td>
+                    <span className={statusClass(c.status)}>{c.status ?? "active"}</span>
                   </td>
-                <td style={{ padding: "1.25rem 1.5rem" }}>
+                  <td style={{ color: "var(--text-secondary)" }}>
-                  <div style={{ display: "flex", flexWrap: "wrap", gap: "6px" }}>
+                    {c.total_tokens.toLocaleString()} tokens / {formatCost(c.total_cost_usd)}
-                    {c.agents.map(a => (
-                      <span key={a} style={{ fontSize: "0.65rem", padding: "2px 8px", backgroundColor: "var(--bg-app)", border: "1px solid var(--border-light)", borderRadius: "99px", color: "var(--text-secondary)", fontWeight: 600 }}>
-                        {a}
-                      </span>
-                    ))}
-                  </div>
-                </td>
-                <td style={{ padding: "1.25rem 1.5rem" }}>
-                  <span style={{ 
-                    fontSize: "0.75rem", 
-                    padding: "4px 10px", 
-                    borderRadius: "6px", 
-                    fontWeight: 600,
-                    backgroundColor: c.status === "Resolved" ? "#DEF7EC" : "#FDE8E8",
-                    color: c.status === "Resolved" ? "#03543F" : "#9B1C1C",
-                  }}>
-                    {c.status}
-                  </span>
-                  {c.hitl && <span style={{ marginLeft: "8px", fontSize: "1.25rem" }} title="Human in the loop invoked">🔒</span>}
-                </td>
-                <td style={{ padding: "1.25rem 1.5rem", fontSize: "0.875rem", color: "var(--text-secondary)" }}>
-                  {c.turns} turns • {c.cost}
                   </td>
                 </tr>
               ))}
             </tbody>
           </table>
 
-        <div style={{ padding: "1.25rem 1.5rem", borderTop: "1px solid var(--border-light)", display: "flex", justifyContent: "space-between", alignItems: "center", backgroundColor: "var(--bg-surface-inner)" }}>
+          <div className="pagination-bar">
-          <span style={{ fontSize: "0.875rem", color: "var(--text-secondary)" }}>Showing 1-5 of 120 sessions</span>
+            <span className="pagination-bar__info">
-          <div style={{ display: "flex", gap: "0.5rem" }}>
+              Showing {(page - 1) * perPage + 1}-{Math.min(page * perPage, total)} of {total} sessions
+            </span>
+            <div className="pagination-bar__controls">
               <button
                 onClick={(e) => { e.stopPropagation(); setPage(p => Math.max(1, p - 1)) }}
                 disabled={page === 1}
@@ -120,11 +126,7 @@ export function ReplayListPage() {
             </div>
           </div>
         </div>
-      <style>{`
+      )}
-        .replay-row-hover:hover {
-          background-color: var(--bg-hover) !important;
-        }
-      `}</style>
     </div>
   );
 }

frontend/src/pages/ReplayPage.tsx

@@ -1,70 +1,90 @@
-import { useState } from "react";
+import { useState, useEffect } from "react";
 import { useParams, useNavigate } from "react-router-dom";
 import { ReplayTimeline } from "../components/ReplayTimeline";
-
+import { fetchReplay, ReplayStep } from "../api";
-const MOCK_STEPS = [
-  { step: 1, type: "message", timestamp: "2026-04-05T10:00:00Z", agent: "Customer", content: "My laptop arrived with a shattered screen. I need a replacement immediately! Order #8921." },
-  { step: 2, type: "token", timestamp: "2026-04-05T10:00:02Z", agent: "Router", content: "Intent detected: 'return_request'. Routing to Order Specialist." },
-  { step: 3, type: "tool_call", timestamp: "2026-04-05T10:00:03Z", agent: "Order Specialist", tool: "get_order_details", params: { order_id: "8921" } },
-  { step: 4, type: "tool_result", timestamp: "2026-04-05T10:00:04Z", tool: "get_order_details", result: { status: "Delivered", items: ["MacBook Pro 16", "USB-C Hub"], total_value: 2499.00 } },
-  { step: 5, type: "tool_call", timestamp: "2026-04-05T10:00:06Z", agent: "Order Specialist", tool: "initiate_return", params: { order_id: "8921", reason: "Damaged in transit", replacement: true } },
-  { step: 6, type: "interrupt", timestamp: "2026-04-05T10:00:06Z", agent: "System", content: "SECURITY POLICY TRIGGERED: High-Value Return (>$1000). Human approval required before initiating RMS workflow." },
-  { step: 7, type: "interrupt_response", timestamp: "2026-04-05T10:15:22Z", agent: "Alex Thompson (Supervisor)", content: "REJECTED. Standard policy for shattered screens requires photo evidence before dispatching replacement unit." },
-  { step: 8, type: "message", timestamp: "2026-04-05T10:15:25Z", agent: "Order Specialist", content: "I'm so sorry to hear your laptop screen was shattered! Because this is a high-value item, our policy requires a photo of the damage before we can dispatch your replacement unit. Could you please take a quick picture and upload it here?" }
-];
 
 export function ReplayPage() {
   const { threadId } = useParams<{ threadId: string }>();
   const navigate = useNavigate();
-  const [page, setPage] = useState(1);
+  const [steps, setSteps] = useState<ReplayStep[]>([]);
+  const [totalSteps, setTotalSteps] = useState(0);
+  const [isLoading, setIsLoading] = useState(true);
+  const [error, setError] = useState<string | null>(null);
+
+  useEffect(() => {
+    if (!threadId) return;
+    setIsLoading(true);
+    setError(null);
+    fetchReplay(threadId, 1, 100)
+      .then((result) => {
+        setSteps(result.steps);
+        setTotalSteps(result.total_steps);
+      })
+      .catch((err: Error) => setError(err.message))
+      .finally(() => setIsLoading(false));
+  }, [threadId]);
 
   if (!threadId) return null;
 
   return (
     <div className="page-container">
-      <div className="page-header" style={{ display: "flex", justifyContent: "space-between", alignItems: "flex-end", marginBottom: "2rem" }}>
+      <div className="page-header" style={{ marginBottom: "2rem" }}>
-        <div>
         <button
           onClick={() => navigate("/replay")}
           style={{ background: "none", border: "none", color: "var(--text-secondary)", fontSize: "0.875rem", cursor: "pointer", padding: "0 0 0.5rem 0", display: "flex", alignItems: "center", gap: "0.25rem" }}
         >
-            ← Back to All Replays
+          &larr; Back to All Replays
         </button>
         <h2>Audit Trail: <span style={{ fontFamily: "monospace", color: "var(--brand-primary)" }}>{threadId}</span></h2>
         <p>Detailed temporal log of agent reflections, MCP tool calls, and human overrides.</p>
       </div>
-      </div>
 
+      {error ? (
+        <div className="error-state">
+          <p className="error-state__title">Failed to load replay</p>
+          <p className="error-state__description">{error}</p>
+        </div>
+      ) : isLoading ? (
+        <div style={{ display: "grid", gridTemplateColumns: "1fr 3fr", gap: "2rem" }}>
+          <div className="skeleton-box" style={{ height: "250px", borderRadius: "var(--radius-xl)", background: "var(--bg-surface)" }}></div>
+          <div className="skeleton-box" style={{ height: "400px", borderRadius: "var(--radius-xl)", background: "var(--bg-surface)" }}></div>
+        </div>
+      ) : steps.length === 0 ? (
+        <div className="empty-state">
+          <p className="empty-state__title">No replay steps found</p>
+          <p className="empty-state__description">This conversation has no recorded checkpoints.</p>
+        </div>
+      ) : (
         <div style={{ display: "grid", gridTemplateColumns: "1fr 3fr", gap: "2rem" }}>
           {/* Sidebar Summary Info */}
-        <div style={{ backgroundColor: "var(--bg-surface)", padding: "1.5rem", borderRadius: "var(--radius-xl)", border: "1px solid var(--border-light)", alignSelf: "start" }}>
+          <div className="section-card" style={{ alignSelf: "start" }}>
             <h3 style={{ fontSize: "1rem", marginBottom: "1.25rem", color: "var(--text-primary)" }}>Session Context</h3>
-          
             <div style={{ display: "flex", flexDirection: "column", gap: "1rem" }}>
               <div>
-              <div style={{ fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Customer</div>
+                <div className="stat-label">Thread ID</div>
-              <div style={{ fontWeight: 600, fontSize: "0.9375rem" }}>Maria G.</div>
+                <div className="stat-value" style={{ fontSize: "0.8125rem", fontFamily: "monospace", wordBreak: "break-all" }}>{threadId}</div>
               </div>
               <div>
-              <div style={{ fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Final Outcome</div>
+                <div className="stat-label">Total Steps</div>
-              <div style={{ display: "inline-block", backgroundColor: "#FDE8E8", color: "#9B1C1C", padding: "4px 8px", borderRadius: "6px", fontSize: "0.75rem", fontWeight: 700, marginTop: "4px" }}>ESCALATED 🔒</div>
+                <div className="stat-value">{totalSteps}</div>
               </div>
               <div>
-              <div style={{ fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Time Elapsed</div>
+                <div className="stat-label">Time Range</div>
-              <div style={{ fontSize: "0.9375rem" }}>15m 25s</div>
+                <div style={{ fontSize: "0.8125rem" }}>
+                  {steps[0]?.timestamp ? new Date(steps[0].timestamp).toLocaleString() : "N/A"}
+                  {" \u2013 "}
+                  {steps[steps.length - 1]?.timestamp ? new Date(steps[steps.length - 1].timestamp).toLocaleString() : "N/A"}
                 </div>
-            <div>
-              <div style={{ fontSize: "0.75rem", textTransform: "uppercase", color: "var(--text-secondary)", fontWeight: 600 }}>Total Tokens</div>
-              <div style={{ fontSize: "0.9375rem" }}>3,402 ($0.15)</div>
               </div>
             </div>
           </div>
 
           {/* Timeline */}
-        <div style={{ backgroundColor: "var(--bg-surface)", padding: "2rem", borderRadius: "var(--radius-xl)", border: "1px solid var(--border-light)" }}>
+          <div className="section-card" style={{ padding: "2rem" }}>
-          <ReplayTimeline steps={MOCK_STEPS as any} />
+            <ReplayTimeline steps={steps as any} />
           </div>
         </div>
+      )}
     </div>
   );
 }

frontend/src/pages/ReviewPage.tsx

@@ -2,8 +2,8 @@ import { useEffect, useRef, useState } from "react";
 
 interface ImportJob {
   job_id: string;
-  status: "pending" | "processing" | "done" | "error";
+  status: "pending" | "processing" | "done" | "failed";
-  error?: string;
+  error_message?: string;
 }
 
 interface EndpointClassification {
@@ -14,16 +14,9 @@ interface EndpointClassification {
   agent_group: string;
 }
 
-interface JobResult {
-  job_id: string;
-  status: string;
-  endpoints: EndpointClassification[];
-}
-
 export function ReviewPage() {
   const [url, setUrl] = useState("");
   const [job, setJob] = useState<ImportJob | null>(null);
-  const [result, setResult] = useState<JobResult | null>(null);
   const [submitting, setSubmitting] = useState(false);
   const [submitError, setSubmitError] = useState<string | null>(null);
   const [classifications, setClassifications] = useState<EndpointClassification[]>([
@@ -45,7 +38,7 @@ export function ReviewPage() {
       path: "/api/v1/payments/{charge_id}/refund",
       method: "post",
       summary: "Issue a full or partial refund for a charge",
-      access_type: "admin",
+      access_type: "write",
       agent_group: "Billing Assistant",
     },
     {
@@ -78,14 +71,20 @@ export function ReviewPage() {
         const j: ImportJob = data.data ?? data;
         setJob(j);
         if (j.status === "done") {
-          return fetch(`/api/openapi/jobs/${encodeURIComponent(jobId)}/result`)
+          return fetch(`/api/openapi/jobs/${encodeURIComponent(jobId)}/classifications`)
             .then((r) => r.json())
-            .then((rdata) => {
+            .then((clfs: EndpointClassification[]) => {
-              const res: JobResult = rdata.data ?? rdata;
+              setClassifications(
-              setResult(res);
+                clfs.map((c: any) => ({
-              setClassifications(res.endpoints ?? []);
+                  path: c.endpoint?.path ?? c.path ?? "",
+                  method: c.endpoint?.method ?? c.method ?? "",
+                  summary: c.endpoint?.summary ?? c.summary ?? "",
+                  access_type: c.access_type ?? "read",
+                  agent_group: c.agent_group ?? "Unassigned",
+                }))
+              );
             });
-        } else if (j.status === "error") {
+        } else if (j.status === "failed") {
           return;
         } else {
           pollRef.current = setTimeout(() => pollJob(jobId), 2000);
@@ -102,7 +101,6 @@ export function ReviewPage() {
     setSubmitting(true);
     setSubmitError(null);
     setJob(null);
-    setResult(null);
     setClassifications([]);
 
     fetch("/api/openapi/import", {
@@ -174,10 +172,10 @@ export function ReviewPage() {
       {job && (
         <div style={{ padding: "1rem", background: "var(--bg-surface)", border: "1px solid var(--border-light)", borderRadius: "var(--radius-md)", marginBottom: "1.5rem" }}>
           <strong>Job:</strong> {job.job_id} &mdash; Status:{" "}
-          <span style={{ fontWeight: 600, color: job.status === "done" ? "#10b981" : job.status === "error" ? "var(--brand-accent)" : "#f59e0b" }}>
+          <span style={{ fontWeight: 600, color: job.status === "done" ? "#10b981" : job.status === "failed" ? "var(--brand-accent)" : "#f59e0b" }}>
             {job.status}
           </span>
-          {job.error && <div style={{ marginTop: "4px", color: "var(--brand-accent)" }}>{job.error}</div>}
+          {job.error_message && <div style={{ marginTop: "4px", color: "var(--brand-accent)" }}>{job.error_message}</div>}
         </div>
       )}
 
@@ -221,7 +219,6 @@ export function ReviewPage() {
                         >
                           <option value="read">Read Only</option>
                           <option value="write">Write (Confirm)</option>
-                          <option value="admin">Admin</option>
                         </select>
                         <input
                           type="text"

frontend/tsconfig.tsbuildinfo

@@ -1 +1 @@
-{"root":["./src/app.tsx","./src/api.ts","./src/main.tsx","./src/types.ts","./src/components/agentaction.tsx","./src/components/chatinput.tsx","./src/components/chatmessages.tsx","./src/components/errorbanner.tsx","./src/components/interruptprompt.tsx","./src/components/layout.tsx","./src/components/metriccard.tsx","./src/components/navbar.tsx","./src/components/replaytimeline.tsx","./src/hooks/usewebsocket.ts","./src/pages/chatpage.tsx","./src/pages/dashboardpage.tsx","./src/pages/replaylistpage.tsx","./src/pages/replaypage.tsx","./src/pages/reviewpage.tsx"],"version":"5.7.3"}
+{"root":["./src/app.tsx","./src/api.ts","./src/main.tsx","./src/types.ts","./src/vite-env.d.ts","./src/components/agentaction.tsx","./src/components/chatinput.tsx","./src/components/chatmessages.tsx","./src/components/errorbanner.tsx","./src/components/interruptprompt.tsx","./src/components/layout.tsx","./src/components/metriccard.tsx","./src/components/navbar.tsx","./src/components/replaytimeline.tsx","./src/hooks/usewebsocket.ts","./src/pages/chatpage.tsx","./src/pages/dashboardpage.tsx","./src/pages/replaylistpage.tsx","./src/pages/replaypage.tsx","./src/pages/reviewpage.tsx"],"version":"5.7.3"}