@oneuptime/common 11.5.4 → 11.5.6

package/Tests/Server/Utils/AI/CodeAgentWorkspaceGuard.test.ts

@@ -0,0 +1,138 @@
+import CodeAgentWorkspaceGuard, {
+  MAX_TOOL_OUTPUT_CHARS,
+} from "../../../../Server/Utils/AI/CodeFix/CodeAgentWorkspaceGuard";
+import path from "path";
+import { describe, expect, test } from "@jest/globals";
+
+/*
+ * The in-house code agent's pure safety edges (B4 Tier 0): every workspace
+ * tool path must resolve INSIDE the run's ephemeral workspace, and every
+ * tool output must be truncated before it re-enters the model context.
+ * These are the invariants the model itself is not trusted with.
+ */
+
+const workspaceRoot: string = path.join(path.sep, "tmp", "oneuptime-ws");
+
+describe("CodeAgentWorkspaceGuard.resolveWorkspacePath", () => {
+  test("resolves a relative path inside the workspace", () => {
+    expect(
+      CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        "src/billing/charge.ts",
+      ),
+    ).toBe(path.join(workspaceRoot, "src", "billing", "charge.ts"));
+  });
+
+  test("'.' resolves to the workspace root itself", () => {
+    expect(
+      CodeAgentWorkspaceGuard.resolveWorkspacePath(workspaceRoot, "."),
+    ).toBe(workspaceRoot);
+  });
+
+  test("normalizes internal '..' segments that stay inside", () => {
+    expect(
+      CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        "src/../package.json",
+      ),
+    ).toBe(path.join(workspaceRoot, "package.json"));
+  });
+
+  test("refuses '..' traversal out of the workspace", () => {
+    expect(() => {
+      return CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        "../etc/passwd",
+      );
+    }).toThrow("escapes the workspace");
+  });
+
+  test("refuses deep traversal that leaves and re-enters elsewhere", () => {
+    expect(() => {
+      return CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        "src/../../other-workspace/file.ts",
+      );
+    }).toThrow("escapes the workspace");
+  });
+
+  test("allows an absolute path INSIDE the workspace (models echo them back)", () => {
+    const inside: string = path.join(workspaceRoot, "README.md");
+    expect(
+      CodeAgentWorkspaceGuard.resolveWorkspacePath(workspaceRoot, inside),
+    ).toBe(inside);
+  });
+
+  test("refuses an absolute path outside the workspace", () => {
+    expect(() => {
+      return CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        path.join(path.sep, "etc", "passwd"),
+      );
+    }).toThrow("escapes the workspace");
+  });
+
+  test("refuses the sibling-prefix trick (/ws-evil vs /ws)", () => {
+    expect(() => {
+      return CodeAgentWorkspaceGuard.resolveWorkspacePath(
+        workspaceRoot,
+        `${workspaceRoot}-evil/file.ts`,
+      );
+    }).toThrow("escapes the workspace");
+  });
+
+  test("refuses an empty path", () => {
+    expect(() => {
+      return CodeAgentWorkspaceGuard.resolveWorkspacePath(workspaceRoot, "");
+    }).toThrow("file path is required");
+  });
+});
+
+describe("CodeAgentWorkspaceGuard.toWorkspaceRelativePath", () => {
+  test("the root maps to '.'", () => {
+    expect(
+      CodeAgentWorkspaceGuard.toWorkspaceRelativePath(
+        workspaceRoot,
+        workspaceRoot,
+      ),
+    ).toBe(".");
+  });
+
+  test("a nested path maps to its workspace-relative form", () => {
+    expect(
+      CodeAgentWorkspaceGuard.toWorkspaceRelativePath(
+        workspaceRoot,
+        path.join(workspaceRoot, "src", "index.ts"),
+      ),
+    ).toBe(path.join("src", "index.ts"));
+  });
+});
+
+describe("CodeAgentWorkspaceGuard.truncateToolOutput", () => {
+  test("output under the cap is returned unchanged", () => {
+    expect(CodeAgentWorkspaceGuard.truncateToolOutput("short output")).toBe(
+      "short output",
+    );
+  });
+
+  test("output exactly at the cap is returned unchanged", () => {
+    const exact: string = "a".repeat(MAX_TOOL_OUTPUT_CHARS);
+    expect(CodeAgentWorkspaceGuard.truncateToolOutput(exact)).toBe(exact);
+  });
+
+  test("output over the cap is truncated with an explicit marker", () => {
+    const over: string = "a".repeat(MAX_TOOL_OUTPUT_CHARS + 1);
+    const truncated: string = CodeAgentWorkspaceGuard.truncateToolOutput(over);
+
+    expect(truncated.startsWith("a".repeat(MAX_TOOL_OUTPUT_CHARS))).toBe(true);
+    expect(truncated).toContain("[output truncated to");
+    // The marker is bounded — the result cannot balloon past cap + marker.
+    expect(truncated.length).toBeLessThan(MAX_TOOL_OUTPUT_CHARS + 100);
+  });
+
+  test("a custom cap is honored", () => {
+    expect(CodeAgentWorkspaceGuard.truncateToolOutput("abcdef", 3)).toContain(
+      "abc\n... [output truncated to 3 characters]",
+    );
+  });
+});

package/Tests/Server/Utils/AI/CodeFixAgentCompletion.test.ts

@@ -0,0 +1,335 @@
+import CodeFixAgentCompletion, {
+  AgentCompletionResult,
+  LoopBudgetDecision,
+  MAX_COMPLETION_CALLS_PER_RUN,
+  MAX_OUTPUT_TOKENS_PER_RUN,
+  MAX_TOKENS_PER_COMPLETION_CALL,
+} from "../../../../Server/Utils/AI/CodeFix/CodeFixAgentCompletion";
+import AIRunService from "../../../../Server/Services/AIRunService";
+import LlmLogService from "../../../../Server/Services/LlmLogService";
+import LlmProviderService from "../../../../Server/Services/LlmProviderService";
+import AIService, {
+  AILogRequest,
+  AI_CODE_FIX_FEATURE,
+} from "../../../../Server/Services/AIService";
+import AIRun from "../../../../Models/DatabaseModels/AIRun";
+import LlmProvider from "../../../../Models/DatabaseModels/LlmProvider";
+import LlmLog from "../../../../Models/DatabaseModels/LlmLog";
+import AIRunStatus from "../../../../Types/AI/AIRunStatus";
+import AIRunType from "../../../../Types/AI/AIRunType";
+import ObjectID from "../../../../Types/ObjectID";
+import { LLMMessage } from "../../../../Server/Utils/LLM/LLMService";
+import { describe, expect, test, afterEach, beforeEach } from "@jest/globals";
+
+/*
+ * The server-mediated code-fix completion endpoint (B4 Tier 0). These tests
+ * lock in the guard matrix the design doc requires: completions are served
+ * ONLY for a claimed Running CodeFix run owned by the calling agent, per-run
+ * loop budgets are enforced server-side (never trusted to the worker), the
+ * provider comes from the METERED resolution, and the call executes under
+ * the "AI Code Fix" feature with content previews redacted.
+ */
+
+const agentId: ObjectID = ObjectID.generate();
+const runId: ObjectID = ObjectID.generate();
+const projectId: ObjectID = ObjectID.generate();
+
+const messages: Array<LLMMessage> = [
+  { role: "system", content: "recipe prompt" },
+  { role: "user", content: "begin" },
+];
+
+function fakeRun(overrides: Partial<AIRun> = {}): AIRun {
+  return {
+    id: runId,
+    projectId,
+    runType: AIRunType.CodeFix,
+    status: AIRunStatus.Running,
+    aiAgentId: agentId,
+    ...overrides,
+  } as unknown as AIRun;
+}
+
+function fakeProvider(): LlmProvider {
+  return {
+    id: ObjectID.generate(),
+    name: "provider",
+    isGlobalLlm: true,
+  } as unknown as LlmProvider;
+}
+
+type ExecuteSpy = jest.SpiedFunction<typeof AIService.executeWithLogging>;
+type UpdateSpy = jest.SpiedFunction<typeof AIRunService.updateOneBy>;
+
+function mockHappyDependencies(data?: {
+  completionCalls?: number;
+  outputTokens?: number;
+}): { executeSpy: ExecuteSpy; updateSpy: UpdateSpy } {
+  jest.spyOn(AIRunService, "findOneById").mockResolvedValue(fakeRun());
+  const updateSpy: UpdateSpy = jest
+    .spyOn(AIRunService, "updateOneBy")
+    .mockResolvedValue(1);
+  jest.spyOn(LlmLogService, "getRunCompletionUsage").mockResolvedValue({
+    completionCalls: data?.completionCalls ?? 2,
+    outputTokens: data?.outputTokens ?? 500,
+  });
+  jest
+    .spyOn(LlmProviderService, "getLlmProviderForMeteredAgentPath")
+    .mockResolvedValue(fakeProvider());
+
+  const executeSpy: ExecuteSpy = jest
+    .spyOn(AIService, "executeWithLogging")
+    .mockResolvedValue({
+      content: "done",
+      toolCalls: undefined,
+      llmLog: { completionTokens: 123 } as unknown as LlmLog,
+    });
+
+  return { executeSpy, updateSpy };
+}
+
+describe("CodeFixAgentCompletion.evaluateLoopBudget (pure)", () => {
+  test("allows a run under both budgets", () => {
+    const decision: LoopBudgetDecision =
+      CodeFixAgentCompletion.evaluateLoopBudget({
+        completionCalls: MAX_COMPLETION_CALLS_PER_RUN - 1,
+        outputTokens: MAX_OUTPUT_TOKENS_PER_RUN - 1,
+      });
+
+    expect(decision.allowed).toBe(true);
+    expect(decision.reason).toBeNull();
+  });
+
+  test("refuses at the completion-call cap, naming the numbers", () => {
+    const decision: LoopBudgetDecision =
+      CodeFixAgentCompletion.evaluateLoopBudget({
+        completionCalls: MAX_COMPLETION_CALLS_PER_RUN,
+        outputTokens: 0,
+      });
+
+    expect(decision.allowed).toBe(false);
+    expect(decision.reason).toContain(`${MAX_COMPLETION_CALLS_PER_RUN}`);
+  });
+
+  test("refuses at the output-token cap, naming the numbers", () => {
+    const decision: LoopBudgetDecision =
+      CodeFixAgentCompletion.evaluateLoopBudget({
+        completionCalls: 0,
+        outputTokens: MAX_OUTPUT_TOKENS_PER_RUN,
+      });
+
+    expect(decision.allowed).toBe(false);
+    expect(decision.reason).toContain("output-token");
+  });
+});
+
+describe("CodeFixAgentCompletion.execute guards", () => {
+  beforeEach(() => {
+    // Fails the test loudly if a guard leaks through to execution.
+    jest.spyOn(AIService, "executeWithLogging").mockImplementation(() => {
+      throw new Error("executeWithLogging must not be reached");
+    });
+  });
+
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("refuses empty messages", async () => {
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages: [],
+      }),
+    ).rejects.toThrow("messages must be a non-empty array");
+  });
+
+  test("refuses when the run does not exist", async () => {
+    jest.spyOn(AIRunService, "findOneById").mockResolvedValue(null);
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("Task not found");
+  });
+
+  test("refuses a non-CodeFix run", async () => {
+    jest
+      .spyOn(AIRunService, "findOneById")
+      .mockResolvedValue(fakeRun({ runType: AIRunType.Investigation }));
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("only served for code-fix runs");
+  });
+
+  test("refuses a run that is not Running (terminal or queued)", async () => {
+    jest
+      .spyOn(AIRunService, "findOneById")
+      .mockResolvedValue(fakeRun({ status: AIRunStatus.Completed }));
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("not running");
+  });
+
+  test("refuses a run claimed by a DIFFERENT agent", async () => {
+    jest
+      .spyOn(AIRunService, "findOneById")
+      .mockResolvedValue(fakeRun({ aiAgentId: ObjectID.generate() }));
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("claimed by a different agent");
+  });
+
+  test("refuses an over-budget run (call cap) without executing", async () => {
+    jest.spyOn(AIRunService, "findOneById").mockResolvedValue(fakeRun());
+    jest.spyOn(LlmLogService, "getRunCompletionUsage").mockResolvedValue({
+      completionCalls: MAX_COMPLETION_CALLS_PER_RUN,
+      outputTokens: 0,
+    });
+    const providerSpy: jest.SpyInstance = jest.spyOn(
+      LlmProviderService,
+      "getLlmProviderForMeteredAgentPath",
+    );
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("LLM call budget");
+
+    // Over-budget short-circuits before provider resolution and execution.
+    expect(providerSpy).not.toHaveBeenCalled();
+  });
+
+  test("refuses an over-budget run (output tokens) without executing", async () => {
+    jest.spyOn(AIRunService, "findOneById").mockResolvedValue(fakeRun());
+    jest.spyOn(LlmLogService, "getRunCompletionUsage").mockResolvedValue({
+      completionCalls: 1,
+      outputTokens: MAX_OUTPUT_TOKENS_PER_RUN,
+    });
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("output-token budget");
+  });
+
+  test("refuses when no provider resolves on the metered path", async () => {
+    jest.spyOn(AIRunService, "findOneById").mockResolvedValue(fakeRun());
+    jest.spyOn(LlmLogService, "getRunCompletionUsage").mockResolvedValue({
+      completionCalls: 0,
+      outputTokens: 0,
+    });
+    jest
+      .spyOn(LlmProviderService, "getLlmProviderForMeteredAgentPath")
+      .mockResolvedValue(null);
+
+    await expect(
+      CodeFixAgentCompletion.execute({
+        aiAgentId: agentId,
+        aiRunId: runId,
+        messages,
+      }),
+    ).rejects.toThrow("No LLM provider is available");
+  });
+});
+
+describe("CodeFixAgentCompletion.execute happy path", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("executes under the AI Code Fix feature, run-linked and preview-redacted", async () => {
+    const { executeSpy } = mockHappyDependencies();
+
+    const result: AgentCompletionResult = await CodeFixAgentCompletion.execute({
+      aiAgentId: agentId,
+      aiRunId: runId,
+      messages,
+    });
+
+    expect(executeSpy).toHaveBeenCalledTimes(1);
+    const request: AILogRequest = executeSpy.mock.calls[0]![0]!;
+    expect(request.feature).toBe(AI_CODE_FIX_FEATURE);
+    expect(request.aiRunId).toBe(runId);
+    expect(request.projectId).toBe(projectId);
+    expect(request.storeContentPreviews).toBe(false);
+    // No worker-supplied maxTokens => the per-call cap applies.
+    expect(request.maxTokens).toBe(MAX_TOKENS_PER_COMPLETION_CALL);
+
+    expect(result.content).toBe("done");
+    expect(result.toolCalls).toEqual([]);
+    expect(result.stopReason).toBe("stop");
+  });
+
+  test("clamps a worker-supplied maxTokens to the per-call cap", async () => {
+    const { executeSpy } = mockHappyDependencies();
+
+    await CodeFixAgentCompletion.execute({
+      aiAgentId: agentId,
+      aiRunId: runId,
+      messages,
+      maxTokens: 999_999,
+    });
+
+    expect(executeSpy.mock.calls[0]![0]!.maxTokens).toBe(
+      MAX_TOKENS_PER_COMPLETION_CALL,
+    );
+  });
+
+  test("reports post-call budget usage so the worker can wind down early", async () => {
+    mockHappyDependencies({ completionCalls: 5, outputTokens: 1000 });
+
+    const result: AgentCompletionResult = await CodeFixAgentCompletion.execute({
+      aiAgentId: agentId,
+      aiRunId: runId,
+      messages,
+    });
+
+    expect(result.budget).toEqual({
+      completionCallsUsed: 6,
+      maxCompletionCalls: MAX_COMPLETION_CALLS_PER_RUN,
+      outputTokensUsed: 1000 + 123,
+      maxOutputTokens: MAX_OUTPUT_TOKENS_PER_RUN,
+    });
+  });
+
+  test("a completion refreshes the run heartbeat (Running-guarded)", async () => {
+    const { updateSpy } = mockHappyDependencies();
+
+    await CodeFixAgentCompletion.execute({
+      aiAgentId: agentId,
+      aiRunId: runId,
+      messages,
+    });
+
+    expect(updateSpy).toHaveBeenCalledTimes(1);
+    expect(updateSpy.mock.calls[0]![0]!.query["status"]).toBe(
+      AIRunStatus.Running,
+    );
+  });
+});

package/Tests/Server/Utils/AI/CodeFixRunQueue.test.ts

@@ -0,0 +1,134 @@
+import CodeFixRunQueue from "../../../../Server/Utils/AI/CodeFix/CodeFixRunQueue";
+import AIRunService from "../../../../Server/Services/AIRunService";
+import AIAgentService from "../../../../Server/Services/AIAgentService";
+import AIRun from "../../../../Models/DatabaseModels/AIRun";
+import AIAgent from "../../../../Models/DatabaseModels/AIAgent";
+import AIRunStatus from "../../../../Types/AI/AIRunStatus";
+import ObjectID from "../../../../Types/ObjectID";
+import { describe, expect, test, afterEach } from "@jest/globals";
+
+/*
+ * Sweeper decisions for CodeFix AIRuns:
+ *
+ *   (a) a heartbeat-stale Running run is finalized as Error, never requeued
+ *       — the external agent may have already pushed a partial fix branch,
+ *       so re-running the fix automatically is not safe (this is the
+ *       deliberate divergence from investigations, which requeue);
+ *   (b) a Queued run that outwaited the timeout is failed only when its
+ *       project has no alive agent — a project with an alive agent just
+ *       has a deep queue and is left alone.
+ */
+
+describe("CodeFixRunQueue.markStaleRunAsError", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("finalizes a heartbeat-stale run as Error via a Running-guarded CAS — no requeue", async () => {
+    const runId: ObjectID = ObjectID.generate();
+    const transition: jest.SpyInstance = jest
+      .spyOn(AIRunService, "attemptStatusTransition")
+      .mockResolvedValue(1);
+
+    await CodeFixRunQueue.markStaleRunAsError({ id: runId });
+
+    expect(transition).toHaveBeenCalledTimes(1);
+    expect(transition).toHaveBeenCalledWith(
+      expect.objectContaining({
+        aiRunId: runId,
+        fromStatus: AIRunStatus.Running,
+        set: expect.objectContaining({
+          status: AIRunStatus.Error,
+          completedAt: expect.any(Date),
+        }),
+      }),
+    );
+    // Never a requeue.
+    expect(transition).not.toHaveBeenCalledWith(
+      expect.objectContaining({
+        set: expect.objectContaining({ status: AIRunStatus.Queued }),
+      }),
+    );
+  });
+});
+
+describe("CodeFixRunQueue.failOrphanedQueuedRuns", () => {
+  afterEach(() => {
+    jest.restoreAllMocks();
+  });
+
+  test("fails only runs in projects with no alive agent", async () => {
+    const projectWithAgent: ObjectID = ObjectID.generate();
+    const projectWithoutAgent: ObjectID = ObjectID.generate();
+    const coveredRunId: ObjectID = ObjectID.generate();
+    const orphanedRunId: ObjectID = ObjectID.generate();
+
+    jest.spyOn(AIRunService, "findBy").mockResolvedValue([
+      { id: coveredRunId, projectId: projectWithAgent },
+      { id: orphanedRunId, projectId: projectWithoutAgent },
+    ] as unknown as Array<AIRun>);
+
+    jest
+      .spyOn(AIAgentService, "getConnectedAIAgentForProject")
+      .mockImplementation((projectId: ObjectID): Promise<AIAgent | null> => {
+        return Promise.resolve(
+          projectId.toString() === projectWithAgent.toString()
+            ? ({ id: ObjectID.generate() } as unknown as AIAgent)
+            : null,
+        );
+      });
+
+    const transition: jest.SpyInstance = jest
+      .spyOn(AIRunService, "attemptStatusTransition")
+      .mockResolvedValue(1);
+
+    await CodeFixRunQueue.failOrphanedQueuedRuns();
+
+    expect(transition).toHaveBeenCalledTimes(1);
+    expect(transition).toHaveBeenCalledWith(
+      expect.objectContaining({
+        aiRunId: orphanedRunId,
+        fromStatus: AIRunStatus.Queued,
+        set: expect.objectContaining({
+          status: AIRunStatus.Error,
+          errorMessage: expect.stringContaining("No AI agent picked this"),
+        }),
+      }),
+    );
+  });
+
+  test("does nothing when no run outwaited the timeout", async () => {
+    jest.spyOn(AIRunService, "findBy").mockResolvedValue([]);
+    const aliveLookup: jest.SpyInstance = jest.spyOn(
+      AIAgentService,
+      "getConnectedAIAgentForProject",
+    );
+    const transition: jest.SpyInstance = jest.spyOn(
+      AIRunService,
+      "attemptStatusTransition",
+    );
+
+    await CodeFixRunQueue.failOrphanedQueuedRuns();
+
+    expect(aliveLookup).not.toHaveBeenCalled();
+    expect(transition).not.toHaveBeenCalled();
+  });
+
+  test("looks the alive agent up once per project, not once per run", async () => {
+    const projectId: ObjectID = ObjectID.generate();
+
+    jest.spyOn(AIRunService, "findBy").mockResolvedValue([
+      { id: ObjectID.generate(), projectId },
+      { id: ObjectID.generate(), projectId },
+    ] as unknown as Array<AIRun>);
+
+    const aliveLookup: jest.SpyInstance = jest
+      .spyOn(AIAgentService, "getConnectedAIAgentForProject")
+      .mockResolvedValue(null);
+    jest.spyOn(AIRunService, "attemptStatusTransition").mockResolvedValue(1);
+
+    await CodeFixRunQueue.failOrphanedQueuedRuns();
+
+    expect(aliveLookup).toHaveBeenCalledTimes(1);
+  });
+});