@stigmer/runner 3.15.3-dev.20260914100347 → 3.15.3-dev.20260916211208

package/src/activities/execute-cursor/__tests__/hermetic/tool-error.test.ts

@@ -0,0 +1,138 @@
+/**
+ * Hermetic golden: an UNGATED tool call that FAILS — a `read` the SDK streams
+ * as `running` and then `error` with the failure text as its result — through
+ * the whole `ExecuteCursor` activity.
+ *
+ * Invariant pinned: the error event folds onto the same row
+ * (the builder's `tool_error` upsert) as FAILED, with the failure text
+ * in `error` and `result` EMPTY, `completedAt` stamped, no approval fields (a
+ * read-only built-in is not gated, and the text is not Cursor's hook-block
+ * copy, so the boundary's #205 pass leaves it alone), and the turn still ends
+ * COMPLETED: a failed read is the model's problem to route around, not the
+ * turn's.
+ *
+ * Moved 2026-09-14 (#1097), one line: until then the accumulator
+ * wrote the failure text to BOTH `error` and `result`. Native writes `error`
+ * alone, as do this harness's own sub-agent rows, and every reader shows
+ * `error || result`; the duplicate's one behavioral reader was the twin
+ * collapse's `carriesOwnChange`, which took a hook-denied shell for a row
+ * carrying its own output. `startedAt` was already
+ * present because the `running` event came first.
+ *
+ * Why this net exists: the one FAILED row among the seventeen Cursor goldens
+ * is the gate's own case (`unattributed-hook-block`); the ordinary tool
+ * failure — the fold #1097 rewrote — had no end-to-end net.
+ *
+ * Regenerate ONLY after a deliberate behavior change:
+ *   npx vitest run src/activities/execute-cursor/__tests__/hermetic -u
+ */
+
+import { afterAll, beforeAll, describe, expect, it, vi } from "vitest";
+import { toJson } from "@bufbuild/protobuf";
+import { AgentExecutionStatusSchema } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/api_pb";
+import { ExecutionPhase, ToolCallStatus } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
+
+vi.mock("@cursor/sdk", async () =>
+  (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSdkModule(),
+);
+vi.mock("../../../../client/stigmer-client.js", async () =>
+  (await import("../../../../__test-utils__/hermetic-activity.js")).hermeticStigmerClientModule(),
+);
+
+import {
+  ScriptedClock,
+  createHermeticEnvironment,
+  type HermeticEnvironment,
+} from "../../../../__test-utils__/hermetic-activity.js";
+import { ScriptedCursorAgent, sdkEvents, step } from "../../__test-utils__/scripted-agent.js";
+import {
+  FIXTURE,
+  SDK_CATALOG,
+  beginCursorScenario,
+  cursorExecutionRecord,
+  runCursorTurn,
+} from "../../__test-utils__/hermetic-cursor.js";
+import { stubRegistryFetch } from "../../../../__test-utils__/model-registry-fixture.js";
+
+const AGENT_ID = "agent-hermetic-toolerror-0001";
+const RUN_ID = "run-hermetic-toolerror-0001";
+const CALL_ID = "call-hermetic-toolerror-read-0001";
+const USER_MESSAGE = "What is in MISSING.md?";
+const TEXT_BEFORE = "Let me read it.";
+const READ_ARGS = { path: "MISSING.md" };
+const READ_ERROR = "ENOENT: no such file or directory, open 'MISSING.md'";
+const TEXT_AFTER = "There is no MISSING.md in this workspace.";
+
+describe("ExecuteCursor hermetic — an ungated tool call that fails", () => {
+  let env: HermeticEnvironment;
+  let registry: ReturnType<typeof stubRegistryFetch>;
+  const clock = new ScriptedClock();
+
+  beforeAll(() => {
+    env = createHermeticEnvironment();
+    registry = stubRegistryFetch();
+    clock.install();
+  });
+
+  afterAll(() => {
+    clock.uninstall();
+    registry.restore();
+    env.dispose();
+  });
+
+  it("folds running -> error into one FAILED row carrying the failure text and completes", async () => {
+    // ── Arrange ──────────────────────────────────────────────────────────────
+    const ev = sdkEvents(AGENT_ID, RUN_ID);
+    const agent = new ScriptedCursorAgent({
+      agentId: AGENT_ID,
+      runIds: [RUN_ID],
+      observeStep: () => clock.tick(),
+      turns: [
+        [
+          step.event(ev.init()),
+          step.event(ev.assistant(TEXT_BEFORE)),
+          step.event(ev.toolCall(CALL_ID, "read", "running", READ_ARGS)),
+          step.event(ev.toolCall(CALL_ID, "read", "error", READ_ARGS, READ_ERROR)),
+          step.event(ev.assistant(TEXT_AFTER)),
+          step.turnEnded({ inputTokens: 1_900, outputTokens: 70, cacheReadTokens: 0, cacheWriteTokens: 0 }),
+          step.finished({ result: TEXT_AFTER, model: { id: FIXTURE.model, params: [] } }),
+        ],
+      ],
+    });
+    const record = cursorExecutionRecord({ message: USER_MESSAGE });
+    const scenario = beginCursorScenario({
+      env,
+      clock,
+      record,
+      sdk: { agents: [agent], catalog: SDK_CATALOG },
+    });
+
+    // ── Act ──────────────────────────────────────────────────────────────────
+    const invocation = await runCursorTurn(scenario);
+
+    // ── Assert: a failed read does not fail the turn ─────────────────────────
+    expect(invocation.outcome.kind).toBe("returned");
+    expect((invocation.outcome as { value: Record<string, unknown> }).value.phase).toBe("EXECUTION_COMPLETED");
+    expect(record.persistedPhases.at(-1)).toBe(ExecutionPhase.EXECUTION_COMPLETED);
+
+    // ── Assert: the one FAILED row ───────────────────────────────────────────
+    const rows = record.toolCalls();
+    expect(rows, "running + error fold into ONE row").toHaveLength(1);
+    const row = rows[0];
+    expect(row.id).toBe(CALL_ID);
+    expect(row.status).toBe(ToolCallStatus.TOOL_CALL_FAILED);
+    expect(row.error, "the failure text is the error").toBe(READ_ERROR);
+    expect(row.result, "and only the error").toBe("");
+    expect(row.requiresApproval, "an ordinary failure is not an approval gate").toBe(false);
+    expect(row.approvalRequestedAt).toBe("");
+    expect(row.startedAt < row.completedAt).toBe(true);
+
+    // ── Assert: hermeticity ──────────────────────────────────────────────────
+    expect(registry.urls.every((u) => u.includes("/model-registry"))).toBe(true);
+
+    // ── Assert: the golden ───────────────────────────────────────────────────
+    const final = record.lastFullStatus!;
+    const json = JSON.stringify(toJson(AgentExecutionStatusSchema, final), null, 2) + "\n";
+    await expect(json).toMatchFileSnapshot("./goldens/tool-error.status.json");
+  });
+});

package/src/activities/execute-cursor/__tests__/hermetic/tool-only-step.test.ts

@@ -0,0 +1,130 @@
+/**
+ * Hermetic golden: a turn whose FIRST event after `init` is a tool call — no
+ * assistant text proposed it — through the whole `ExecuteCursor` activity.
+ *
+ * Invariant pinned: a tool call with no AI message before it in
+ * its scope gets an EMPTY AI message created to carry its row
+ * (the builder's AI-message boundary), and the assistant text
+ * that follows the tool opens a new message. So: AI(`""` + the read row),
+ * AI(A1). This is the same shape the native builder produces for a model turn
+ * that proposes tools without text (`ensureAiMessageForToolCall`), which is
+ * why the canonical builder can serve both.
+ *
+ * Predicted for the swap to the shared builder (2026-09-14): NO move.
+ * The boundary rule ("the scope's current AI message, lazily created empty when
+ * the scope's latest model run had no text") produces exactly this transcript.
+ *
+ * Why this net exists: `tool-call.status.json` pins text→tool→text; nothing
+ * pinned a tool with no proposing text before this net.
+ *
+ * Regenerate ONLY after a deliberate behavior change:
+ *   npx vitest run src/activities/execute-cursor/__tests__/hermetic -u
+ */
+
+import { afterAll, beforeAll, describe, expect, it, vi } from "vitest";
+import { toJson } from "@bufbuild/protobuf";
+import { AgentExecutionStatusSchema } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/api_pb";
+import { ExecutionPhase, MessageType, ToolCallStatus } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
+
+vi.mock("@cursor/sdk", async () =>
+  (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSdkModule(),
+);
+vi.mock("../../../../client/stigmer-client.js", async () =>
+  (await import("../../../../__test-utils__/hermetic-activity.js")).hermeticStigmerClientModule(),
+);
+
+import {
+  ScriptedClock,
+  createHermeticEnvironment,
+  type HermeticEnvironment,
+} from "../../../../__test-utils__/hermetic-activity.js";
+import { ScriptedCursorAgent, sdkEvents, step } from "../../__test-utils__/scripted-agent.js";
+import {
+  FIXTURE,
+  SDK_CATALOG,
+  beginCursorScenario,
+  cursorExecutionRecord,
+  runCursorTurn,
+} from "../../__test-utils__/hermetic-cursor.js";
+import { stubRegistryFetch } from "../../../../__test-utils__/model-registry-fixture.js";
+
+const AGENT_ID = "agent-hermetic-toolonly-0001";
+const RUN_ID = "run-hermetic-toolonly-0001";
+const CALL_ID = "call-hermetic-toolonly-read-0001";
+const USER_MESSAGE = "What is in README.md?";
+const READ_ARGS = { path: "README.md" };
+const READ_RESULT = "# Hermetic\n\nA fixture readme.\n";
+const ASSISTANT_TEXT = "README.md holds a one-line fixture description.";
+
+describe("ExecuteCursor hermetic — a tool call with no proposing text", () => {
+  let env: HermeticEnvironment;
+  let registry: ReturnType<typeof stubRegistryFetch>;
+  const clock = new ScriptedClock();
+
+  beforeAll(() => {
+    env = createHermeticEnvironment();
+    registry = stubRegistryFetch();
+    clock.install();
+  });
+
+  afterAll(() => {
+    clock.uninstall();
+    registry.restore();
+    env.dispose();
+  });
+
+  it("creates an empty AI message for the row and a new one for the text that follows", async () => {
+    // ── Arrange ──────────────────────────────────────────────────────────────
+    const ev = sdkEvents(AGENT_ID, RUN_ID);
+    const agent = new ScriptedCursorAgent({
+      agentId: AGENT_ID,
+      runIds: [RUN_ID],
+      observeStep: () => clock.tick(),
+      turns: [
+        [
+          step.event(ev.init()),
+          step.event(ev.toolCall(CALL_ID, "read", "running", READ_ARGS)),
+          step.event(ev.toolCall(CALL_ID, "read", "completed", READ_ARGS, READ_RESULT)),
+          step.event(ev.assistant(ASSISTANT_TEXT)),
+          step.turnEnded({ inputTokens: 1_800, outputTokens: 60, cacheReadTokens: 0, cacheWriteTokens: 0 }),
+          step.finished({ result: ASSISTANT_TEXT, model: { id: FIXTURE.model, params: [] } }),
+        ],
+      ],
+    });
+    const record = cursorExecutionRecord({ message: USER_MESSAGE });
+    const scenario = beginCursorScenario({
+      env,
+      clock,
+      record,
+      sdk: { agents: [agent], catalog: SDK_CATALOG },
+    });
+
+    // ── Act ──────────────────────────────────────────────────────────────────
+    const invocation = await runCursorTurn(scenario);
+
+    // ── Assert: outcome ──────────────────────────────────────────────────────
+    expect(invocation.outcome.kind).toBe("returned");
+    expect((invocation.outcome as { value: Record<string, unknown> }).value.phase).toBe("EXECUTION_COMPLETED");
+    expect(record.persistedPhases.at(-1)).toBe(ExecutionPhase.EXECUTION_COMPLETED);
+
+    // ── Assert: the transcript ───────────────────────────────────────────────
+    const final = record.lastFullStatus!;
+    expect(
+      final.messages.map((m) => [m.type, m.content, m.toolCalls.map((tc) => tc.id)]),
+      "an empty AI message carries the row; the text after the tool is a new message",
+    ).toEqual([
+      [MessageType.MESSAGE_AI, "", [CALL_ID]],
+      [MessageType.MESSAGE_AI, ASSISTANT_TEXT, []],
+    ]);
+    const row = record.toolCalls()[0];
+    expect(row.status).toBe(ToolCallStatus.TOOL_CALL_COMPLETED);
+    expect(row.result).toBe(READ_RESULT);
+
+    // ── Assert: hermeticity ──────────────────────────────────────────────────
+    expect(registry.urls.every((u) => u.includes("/model-registry"))).toBe(true);
+
+    // ── Assert: the golden ───────────────────────────────────────────────────
+    const json = JSON.stringify(toJson(AgentExecutionStatusSchema, final), null, 2) + "\n";
+    await expect(json).toMatchFileSnapshot("./goldens/tool-only-step.status.json");
+  });
+});

package/src/activities/execute-cursor/__tests__/hermetic/unattributed-hook-block.test.ts

@@ -21,6 +21,18 @@
  * Engine disposition on this arm, today: the handle is `close()`d, not parked
  * (index.ts `enterUnattributedHookBlockFailure`).
  *
+ * Moved 2026-09-14 (#1097), one line: the FAILED row — built from
+ * a single `status: "error"` event with no `running` before it — carries
+ * `startedAt`, equal to its `completedAt`: the instant the runner first
+ * learned of the call. Until then a completed-only event left the field unset,
+ * the one field the canonical builder stamps at every row's creation. This
+ * is the only Cursor golden that scripts a completed-only event (the three
+ * moves predicted for that rule were all this one).
+ *
+ * Moved again the same day (#1097), one line: the FAILED row's
+ * `result` no longer duplicates its `error` — the hook-block text lives in
+ * `error` alone, which is the field the #205 detector reads.
+ *
  * Parent phase rows exercised beyond the earlier scenarios: the gate install's
  * foreign-hook detection (`workspace-setup.ts` `mergeHooks`); the FAILED
  * tool-call row built from a single `status: "error"` event; the boundary's
@@ -45,6 +57,9 @@ import {
 vi.mock("@cursor/sdk", async () =>
   (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSdkModule(),
 );
+vi.mock("@cursor/sdk/sqlite", async () =>
+  (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSqliteModule(),
+);
 vi.mock("../../../../client/stigmer-client.js", async () =>
   (await import("../../../../__test-utils__/hermetic-activity.js")).hermeticStigmerClientModule(),
 );
@@ -80,7 +95,7 @@ const USER_HOOKS_JSON =
   JSON.stringify({ version: 1, hooks: { preToolUse: [{ command: FOREIGN_HOOK_COMMAND }] } }, null, 2) + "\n";
 /**
  * The error text Cursor stamps on a hook-blocked tool call
- * (`message-translator.ts` HOOK_BLOCK_ERROR_MARKERS). The SDK has no structured
+ * (`boundary-rows.ts` HOOK_BLOCK_ERROR_MARKERS). The SDK has no structured
  * "denied by hook" signal; this text is the only stream-side trace.
  */
 const HOOK_BLOCK_ERROR = "blocked by a hook";
@@ -123,7 +138,7 @@ describe("ExecuteCursor hermetic — unattributed hook block (#205)", () => {
           // failed with its generic hook-block text. Our ledger never saw it.
           step.event(ev.toolCall(CALL_ID, "shell", "error", SHELL_ARGS, HOOK_BLOCK_ERROR)),
           step.event(ev.assistant(FINAL_TEXT)),
-          step.turnEnded({ inputTokens: 1_500, outputTokens: 40 }),
+          step.turnEnded({ inputTokens: 1_500, outputTokens: 40, cacheReadTokens: 0, cacheWriteTokens: 0 }),
           step.finished({ result: FINAL_TEXT }),
         ],
       ],

package/src/activities/execute-cursor/__tests__/hermetic/workspace-lock-timeout.test.ts

@@ -13,9 +13,9 @@
  * `Agent.create`, no `harness_state_id` bind, nothing streamed.
  *
  * Why this is the one terminal arm WITHOUT a file golden (owner ruling,
- * 2026-09-11, entry 20260911.03 M0): the wire copy embeds the absolute
+ * 2026-09-11, #1070): the wire copy embeds the absolute
  * workspace path, which is a per-run temp directory and differs between macOS
- * and CI. The S0 driver's rule is to control a volatile source at its origin
+ * and CI. The hermetic driver's rule (#1048) is to control a volatile source at its origin
  * or escalate — never redact — so this scenario pins every byte the path does
  * not touch with explicit assertions and builds the expected message with the
  * SAME error class over the real resolved path.
@@ -35,6 +35,9 @@ import { ExecutionPhase, MessageType } from "@stigmer/protos/ai/stigmer/agentic/
 vi.mock("@cursor/sdk", async () =>
   (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSdkModule(),
 );
+vi.mock("@cursor/sdk/sqlite", async () =>
+  (await import("../../__test-utils__/scripted-sdk.js")).scriptedCursorSqliteModule(),
+);
 vi.mock("../../../../client/stigmer-client.js", async () =>
   (await import("../../../../__test-utils__/hermetic-activity.js")).hermeticStigmerClientModule(),
 );