@stigmer/react 3.0.9-dev.20260616060535 → 3.1.0

package/src/execution/__tests__/blanked-narration.test.tsx

@@ -0,0 +1,149 @@
+import { describe, it, expect, afterEach } from "vitest";
+import { render, cleanup } from "@testing-library/react";
+import { create } from "@bufbuild/protobuf";
+import {
+  AgentExecutionSchema,
+  AgentExecutionStatusSchema,
+  type AgentExecution,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/api_pb";
+import { AgentExecutionSpecSchema } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/spec_pb";
+import { ApiResourceMetadataSchema } from "@stigmer/protos/ai/stigmer/commons/apiresource/metadata_pb";
+import {
+  AgentMessageSchema,
+  ToolCallSchema,
+  type AgentMessage,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/message_pb";
+import {
+  ExecutionPhase,
+  MessageType,
+  ToolCallStatus,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
+import { buildThreadItems } from "../MessageThread";
+import { MessageEntry } from "../MessageEntry";
+
+afterEach(cleanup);
+
+// ---------------------------------------------------------------------------
+// Phase 5 — Cursor deny-only normalization (append-only by construction).
+//
+// When a Cursor turn pauses for approval, the runner BLANKS (clears the content
+// of) the model's provisional post-denial narration in place rather than
+// removing the messages — so the WAITING_FOR_APPROVAL finalize is append-only
+// and the backend needs no shrink exception. The cleanliness then lives in the
+// DATA: the SDK's existing empty-message handling renders the blanked messages
+// as nothing, so no per-harness UI special-casing is required.
+//
+// These tests LOCK that existing handling. If a future refactor "un-hides"
+// empty messages, these fail loudly — the runner's blanked narration must never
+// silently leak back into any surface.
+// ---------------------------------------------------------------------------
+
+function aiText(content: string): AgentMessage {
+  return create(AgentMessageSchema, { type: MessageType.MESSAGE_AI, content });
+}
+
+function thinking(content: string): AgentMessage {
+  return create(AgentMessageSchema, { type: MessageType.MESSAGE_THINKING, content });
+}
+
+function gatedAiMessage(): AgentMessage {
+  return create(AgentMessageSchema, {
+    type: MessageType.MESSAGE_AI,
+    content: "",
+    toolCalls: [
+      create(ToolCallSchema, {
+        id: "c1",
+        name: "edit",
+        status: ToolCallStatus.TOOL_CALL_WAITING_APPROVAL,
+      }),
+    ],
+  });
+}
+
+/**
+ * The exact transcript the runner persists after redaction: pre-tool text, the
+ * gated tool call (WAITING_APPROVAL), then the blanked provisional narration
+ * (a THINKING and an AI message, both with empty content). Count preserved.
+ */
+function waitingForApprovalExecution(): AgentExecution {
+  const exec = create(AgentExecutionSchema);
+  exec.metadata = create(ApiResourceMetadataSchema, { id: "exec-hitl" });
+  exec.spec = create(AgentExecutionSpecSchema, { message: "make the change" });
+  exec.status = create(AgentExecutionStatusSchema, {
+    phase: ExecutionPhase.EXECUTION_WAITING_FOR_APPROVAL,
+    messages: [
+      aiText("Let me create the file."),
+      gatedAiMessage(),
+      thinking(""), // blanked provisional reasoning
+      aiText(""), // blanked provisional verdict
+    ],
+  });
+  return exec;
+}
+
+describe("buildThreadItems — blanked post-denial narration (Phase 5)", () => {
+  it("surfaces the approval tool-group but emits no narration message item", () => {
+    const exec = waitingForApprovalExecution();
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    // The approval surface is driven by the WAITING_APPROVAL tool call — its
+    // tool-group is present.
+    const toolGroups = items.filter((i) => i.kind === "tool-group");
+    expect(toolGroups).toHaveLength(1);
+    expect(toolGroups[0].kind === "tool-group" && toolGroups[0].toolCalls[0].status).toBe(
+      ToolCallStatus.TOOL_CALL_WAITING_APPROVAL,
+    );
+
+    // The pre-tool text survives as the only non-empty AI message item; the
+    // blanked AI verdict is gone.
+    const messageItems = items.filter((i) => i.kind === "message");
+    const nonEmptyAi = messageItems.filter(
+      (i) =>
+        i.kind === "message" &&
+        i.message.type === MessageType.MESSAGE_AI &&
+        i.message.content.trim().length > 0,
+    );
+    expect(nonEmptyAi).toHaveLength(1);
+    expect(nonEmptyAi[0].kind === "message" && nonEmptyAi[0].message.content).toBe(
+      "Let me create the file.",
+    );
+
+    // No message item carries the provisional narration text. The empty AI
+    // verdict is skipped entirely; the empty THINKING (if emitted) carries no
+    // content and renders nothing (asserted below at the render layer).
+    for (const item of messageItems) {
+      if (item.kind !== "message") continue;
+      expect(item.message.content).not.toContain("enable the hook");
+    }
+  });
+
+  it("the empty AI verdict is skipped and the empty THINKING carries no content", () => {
+    const exec = waitingForApprovalExecution();
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+    const messageItems = items.filter((i) => i.kind === "message");
+
+    // The empty AI verdict is dropped from the item list entirely (empty-AI
+    // skip). Any surviving message item (e.g. the empty THINKING) has no
+    // content, so the render layer hides it (asserted below).
+    for (const item of messageItems) {
+      if (item.kind !== "message") continue;
+      if (item.message.type === MessageType.MESSAGE_AI) {
+        expect(item.message.content.trim().length).toBeGreaterThan(0);
+      }
+    }
+  });
+});
+
+describe("MessageEntry — blanked narration renders nothing (Phase 5)", () => {
+  it("renders nothing for a blanked AI narration message", () => {
+    const { container } = render(<MessageEntry message={aiText("")} />);
+    expect(container.textContent?.trim()).toBe("");
+  });
+
+  it("renders nothing for a blanked THINKING narration message", () => {
+    const { container } = render(<MessageEntry message={thinking("")} />);
+    expect(container.innerHTML).toBe("");
+  });
+});

package/src/execution/__tests__/buildThreadItems-approvals.test.ts

@@ -0,0 +1,304 @@
+import { describe, it, expect } from "vitest";
+import { create } from "@bufbuild/protobuf";
+import {
+  AgentExecutionSchema,
+  AgentExecutionStatusSchema,
+  type AgentExecution,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/api_pb";
+import { AgentExecutionSpecSchema } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/spec_pb";
+import { ApiResourceMetadataSchema } from "@stigmer/protos/ai/stigmer/commons/apiresource/metadata_pb";
+import {
+  AgentMessageSchema,
+  ToolCallSchema,
+  type AgentMessage,
+  type ToolCall,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/message_pb";
+import {
+  PendingApprovalSchema,
+  type PendingApproval,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/approval_pb";
+import {
+  SubAgentExecutionSchema,
+  type SubAgentExecution,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/subagent_pb";
+import {
+  ExecutionPhase,
+  MessageType,
+  ToolCallStatus,
+} from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
+import { buildThreadItems, type ThreadItem } from "../MessageThread";
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function toolCall(id: string, name: string): ToolCall {
+  return create(ToolCallSchema, {
+    id,
+    name,
+    status: ToolCallStatus.TOOL_CALL_WAITING_APPROVAL,
+  });
+}
+
+function aiWithTools(toolCalls: ToolCall[]): AgentMessage {
+  return create(AgentMessageSchema, {
+    type: MessageType.MESSAGE_AI,
+    content: "",
+    toolCalls,
+  });
+}
+
+function approval(toolCallId: string, toolName: string): PendingApproval {
+  return create(PendingApprovalSchema, {
+    toolCallId,
+    toolName,
+    argsPreview: "{}",
+  });
+}
+
+function execution(opts: {
+  messages?: AgentMessage[];
+  subAgents?: SubAgentExecution[];
+  pendingApprovals?: PendingApproval[];
+}): AgentExecution {
+  return create(AgentExecutionSchema, {
+    metadata: create(ApiResourceMetadataSchema, { id: "aex-1" }),
+    spec: create(AgentExecutionSpecSchema, { message: "go" }),
+    status: create(AgentExecutionStatusSchema, {
+      phase: ExecutionPhase.EXECUTION_WAITING_FOR_APPROVAL,
+      messages: opts.messages ?? [],
+      subAgentExecutions: opts.subAgents ?? [],
+      pendingApprovals: opts.pendingApprovals ?? [],
+    }),
+  });
+}
+
+function approvalItems(items: ThreadItem[]): ThreadItem[] {
+  return items.filter((i) => i.kind === "approval-request");
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+describe("buildThreadItems — inline vs. bottom approval partition", () => {
+  it("does NOT emit a bottom card for an approval rendered inline on a regular tool", () => {
+    const tc = toolCall("tc-edit", "edit_file");
+    const exec = execution({
+      messages: [aiWithTools([tc])],
+      pendingApprovals: [approval("tc-edit", "edit_file")],
+    });
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    // The tool renders in a group → its gate shows inline → no bottom card.
+    expect(approvalItems(items)).toHaveLength(0);
+    expect(items.some((i) => i.kind === "tool-group")).toBe(true);
+  });
+
+  it("emits a bottom card for an orphan approval with no matching tool row", () => {
+    const tc = toolCall("tc-edit", "edit_file");
+    const exec = execution({
+      messages: [aiWithTools([tc])],
+      pendingApprovals: [approval("tc-ghost", "phantom_tool")],
+    });
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    const bottom = approvalItems(items);
+    expect(bottom).toHaveLength(1);
+    expect(
+      bottom[0].kind === "approval-request" && bottom[0].pendingApproval.toolCallId,
+    ).toBe("tc-ghost");
+  });
+
+  it("emits a bottom card for a task-spawn approval (rendered as a sub-agent, not a tool row)", () => {
+    // A `task` call with a matching SubAgentExecution renders as a SubAgentSection,
+    // not an approval-capable ToolCallItem — so its spawn gate must surface below.
+    const task = toolCall("tc-task", "task");
+    const sub = create(SubAgentExecutionSchema, { id: "tc-task", name: "researcher" });
+    const exec = execution({
+      messages: [aiWithTools([task])],
+      subAgents: [sub],
+      pendingApprovals: [approval("tc-task", "task")],
+    });
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    expect(items.some((i) => i.kind === "sub-agent")).toBe(true);
+    expect(approvalItems(items)).toHaveLength(1);
+  });
+
+  it("does NOT emit a bottom card for an approval on a sub-agent's nested tool", () => {
+    const task = toolCall("tc-task", "task");
+    const sub = create(SubAgentExecutionSchema, {
+      id: "tc-task",
+      name: "researcher",
+      messages: [aiWithTools([toolCall("tc-nested", "fetch")])],
+    });
+    const exec = execution({
+      messages: [aiWithTools([task])],
+      subAgents: [sub],
+      pendingApprovals: [approval("tc-nested", "fetch")],
+    });
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    // The nested tool renders inside the sub-agent section → inline gate → no bottom card.
+    expect(approvalItems(items)).toHaveLength(0);
+  });
+
+  it("emits no approval items at all when approvals are disabled", () => {
+    const tc = toolCall("tc-edit", "edit_file");
+    const exec = execution({
+      messages: [aiWithTools([tc])],
+      pendingApprovals: [approval("tc-ghost", "phantom_tool")],
+    });
+
+    const items = buildThreadItems([exec], null, null, false, undefined);
+
+    expect(approvalItems(items)).toHaveLength(0);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// Triangulation (CLIENT boundary)
+//
+// The runner boundary (hitl-resume-approve-all.test.ts) and the server boundary
+// (Go controller update_status tests) are covered elsewhere. This pins that the
+// SDK thread derivation is a faithful PASSTHROUGH of the persisted transcript:
+// every persisted thinking block and tool call must surface as a thread item,
+// across both the gated (approval pending) state and the decided/leased resume
+// state the user reported. A passing test here exonerates the client and
+// localizes the reported "vanishing thinking + first tool call" upstream (the
+// server merge); a failing test would indict the derivation instead.
+// ---------------------------------------------------------------------------
+
+function thinkingMsg(content: string): AgentMessage {
+  return create(AgentMessageSchema, {
+    type: MessageType.MESSAGE_THINKING,
+    content,
+  });
+}
+
+function aiText(content: string): AgentMessage {
+  return create(AgentMessageSchema, {
+    type: MessageType.MESSAGE_AI,
+    content,
+  });
+}
+
+function completedToolCall(id: string, name: string): ToolCall {
+  return create(ToolCallSchema, {
+    id,
+    name,
+    status: ToolCallStatus.TOOL_CALL_COMPLETED,
+  });
+}
+
+function execAtPhase(
+  phase: ExecutionPhase,
+  messages: AgentMessage[],
+  pendingApprovals: PendingApproval[] = [],
+): AgentExecution {
+  return create(AgentExecutionSchema, {
+    metadata: create(ApiResourceMetadataSchema, { id: "aex-1" }),
+    spec: create(AgentExecutionSpecSchema, { message: "Self-DM me" }),
+    status: create(AgentExecutionStatusSchema, {
+      phase,
+      messages,
+      pendingApprovals,
+    }),
+  });
+}
+
+function hasThinkingItem(items: ThreadItem[], content: string): boolean {
+  return items.some(
+    (i) =>
+      i.kind === "message" &&
+      i.message.type === MessageType.MESSAGE_THINKING &&
+      i.message.content === content,
+  );
+}
+
+function hasToolInGroup(items: ThreadItem[], toolCallId: string): boolean {
+  return items.some(
+    (i) => i.kind === "tool-group" && i.toolCalls.some((tc) => tc.id === toolCallId),
+  );
+}
+
+describe("buildThreadItems — approve-all transcript renders faithfully (client triangulation)", () => {
+  it("renders the leading thinking block and the gated tool while approval is pending", () => {
+    const exec = execAtPhase(
+      ExecutionPhase.EXECUTION_WAITING_FOR_APPROVAL,
+      [
+        thinkingMsg("planning the self-DM"),
+        aiWithTools([toolCall("tc-getappstate", "getAppState")]),
+      ],
+      [approval("tc-getappstate", "getAppState")],
+    );
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    expect(hasThinkingItem(items, "planning the self-DM")).toBe(true);
+    expect(hasToolInGroup(items, "tc-getappstate")).toBe(true);
+  });
+
+  it("renders thinking + first tool + appended leased tools after approve-all resume", () => {
+    const exec = execAtPhase(ExecutionPhase.EXECUTION_IN_PROGRESS, [
+      thinkingMsg("planning the self-DM"),
+      aiWithTools([completedToolCall("tc-getappstate", "getAppState")]),
+      aiWithTools([completedToolCall("tc-click", "click")]),
+      aiText("Captured the app state and clicked the element."),
+    ]);
+
+    const items = buildThreadItems([exec], null, null, true, undefined);
+
+    // The leading thinking block and the FIRST tool call survive rendering...
+    expect(hasThinkingItem(items, "planning the self-DM")).toBe(true);
+    expect(hasToolInGroup(items, "tc-getappstate")).toBe(true);
+    // ...alongside the appended leased tool and the final assistant turn.
+    expect(hasToolInGroup(items, "tc-click")).toBe(true);
+    expect(
+      items.some(
+        (i) =>
+          i.kind === "message" &&
+          i.message.content === "Captured the app state and clicked the element.",
+      ),
+    ).toBe(true);
+  });
+});
+
+// ---------------------------------------------------------------------------
+// Active-execution dedupe (regression for the duplicate-key flake)
+//
+// The hook passes `completedExecutions` (filtered on activeExecutionId) plus
+// `activeStreamExecution` (resolved from stream.execution ??
+// fetchedActiveExecution). A transient skew between those two sources can leave
+// the active execution in BOTH inputs. Without deduping, the same execution is
+// emitted twice → duplicate React keys → React drops one subtree, intermittently
+// the one carrying a live ApprovalCard, so the gate buttons vanish ("0 approvals
+// needed"). buildThreadItems must treat activeStreamExecution as authoritative
+// for its id and emit each execution exactly once.
+// ---------------------------------------------------------------------------
+
+describe("buildThreadItems — active-execution dedupe", () => {
+  it("emits each execution once when the active one is in BOTH inputs (no duplicate keys)", () => {
+    const exec = execution({
+      messages: [aiWithTools([toolCall("tc-edit", "edit_file")])],
+      pendingApprovals: [approval("tc-edit", "edit_file")],
+    });
+
+    // Same execution arrives via the list AND as the active stream copy.
+    const items = buildThreadItems([exec], exec, null, true, undefined);
+
+    const keys = items.map((i) => i.key);
+    expect(new Set(keys).size).toBe(keys.length); // all keys unique
+    // Exactly one synthetic spec message and one tool group survive.
+    expect(items.filter((i) => i.kind === "message")).toHaveLength(1);
+    expect(items.filter((i) => i.kind === "tool-group")).toHaveLength(1);
+    // The inline gate is preserved (not dropped with a duplicated subtree).
+    expect(hasToolInGroup(items, "tc-edit")).toBe(true);
+    expect(approvalItems(items)).toHaveLength(0);
+  });
+});