fix(mcp): declare binary session prerequisite process effects

Recover the isolated effect metadata correction identified in #1669. Keep the generic annotation-algebra fixture independent of operation-specific effects.

Co-authored-by: Pompilio Donato <238757198+pompiliodonatoapp@users.noreply.github.com>

scripts/verify-package-platform.d.mts

@@ -8,4 +8,10 @@ export function assertLinuxPackageProviderFailure(execution: {
 export function verifyPackagePlatform(input: {
   readonly cli: string;
   readonly environment: Readonly<Record<string, string | undefined>>;
+  readonly platform?: string;
+  readonly runCommand?: (
+    command: string,
+    arguments_: readonly string[],
+    environment: Readonly<Record<string, string | undefined>>,
+  ) => Promise<string>;
 }): Promise<void>;

scripts/verify-package-platform.mjs

@@ -1,8 +1,13 @@
 import { json, run, runWithStatus } from "./lib/verify-package-core.mjs";
 
 /** Run platform-specific deep-analysis smoke tests. */
-export async function verifyPackagePlatform({ cli, environment }) {
-  if (process.platform === "linux") {
+export async function verifyPackagePlatform({
+  cli,
+  environment,
+  platform = process.platform,
+  runCommand = run,
+}) {
+  if (platform === "linux") {
     const unsupportedExecution = await runWithStatus(
       cli,
       ["analyze", process.execPath, "--json"],
@@ -11,25 +16,27 @@ export async function verifyPackagePlatform({ cli, environment }) {
     assertLinuxPackageProviderFailure(unsupportedExecution);
   } else {
     const overview = json(
-      await run(cli, ["analyze", process.execPath, "--json"], environment),
+      await runCommand(
+        cli,
+        ["analyze", process.execPath, "--json"],
+        environment,
+      ),
     );
-    if (
-      overview.operation !== "binary_overview" ||
-      overview.normalized_result?.procedure_count < 1
-    )
+    if (!hasPositiveProcedureCount(overview))
       throw new Error(
         `packaged Hopper-backed analyze CLI failed: ${JSON.stringify(overview)}`,
       );
     const inspected = json(
-      await run(cli, ["inspect", process.execPath, "--json"], environment),
+      await runCommand(
+        cli,
+        ["inspect", process.execPath, "--json"],
+        environment,
+      ),
     );
-    if (
-      inspected.operation !== "binary_overview" ||
-      inspected.normalized_result?.procedure_count < 1
-    )
+    if (!hasPositiveProcedureCount(inspected))
       throw new Error("packaged inspect CLI failed");
     const functionResult = json(
-      await run(
+      await runCommand(
         cli,
         ["function", process.execPath, "0x1000", "--json"],
         environment,
@@ -44,7 +51,7 @@ export async function verifyPackagePlatform({ cli, environment }) {
         `packaged function CLI failed: ${JSON.stringify(functionResult)}`,
       );
     const xrefs = json(
-      await run(
+      await runCommand(
         cli,
         ["xrefs", process.execPath, "0x1000", "--json"],
         environment,
@@ -56,7 +63,7 @@ export async function verifyPackagePlatform({ cli, environment }) {
     )
       throw new Error(`packaged xrefs CLI failed: ${JSON.stringify(xrefs)}`);
     const trace = json(
-      await run(
+      await runCommand(
         cli,
         ["trace", process.execPath, "fixture", "--json"],
         environment,
@@ -93,3 +100,10 @@ export const assertLinuxPackageProviderFailure = (execution) => {
       `packaged Linux Hopper verification did not fail closed: ${JSON.stringify(failure)}`,
     );
 };
+
+const hasPositiveProcedureCount = (result) =>
+  typeof result === "object" &&
+  result !== null &&
+  result.operation === "binary_overview" &&
+  Number.isInteger(result.normalized_result?.procedure_count) &&
+  result.normalized_result.procedure_count > 0;

src/browser/CdpBrowserProvider.ts

@@ -43,10 +43,7 @@ import {
   type WebScreenshotDiff,
 } from "../domain/webScreenshot.js";
 import { AnalysisError } from "../domain/analysisErrorBase.js";
-import {
-  AnalysisCancelledError,
-  AnalysisInputError,
-} from "../domain/analysisErrorCore.js";
+import { AnalysisInputError } from "../domain/analysisErrorCore.js";
 import { BrowserObservationError } from "../domain/browserObservationError.js";
 import { ProviderAdapterError } from "../domain/providerAdapterError.js";
 import type { BrowserObservationOperation } from "../domain/browserObservationErrors.js";
@@ -167,8 +164,6 @@ export class CdpBrowserProvider implements BrowserObservationPort {
         ),
       );
     } catch (cause: unknown) {
-      if (options.signal?.aborted === true)
-        return err(new AnalysisCancelledError("analyze_web_bundle", { cause }));
       return err(providerError(cause, "analyze_web_bundle"));
     }
   }
@@ -340,10 +335,6 @@ const withTargetSession = async <
       );
     // An unadopted browser socket is closed here; closing is idempotent.
     await endpoint.connection?.close();
-    // Cleanup can observe an interrupt after work returns. Do not publish its
-    // result once cancellation arrived while the target session was closing.
-    if (options.signal?.aborted === true)
-      throw new AnalysisCancelledError(operation);
   }
 };
 

src/contracts/toolEffects.test.ts

@@ -1,7 +1,11 @@
 import { describe, expect, it } from "vitest";
 
 import { TOOL_CONTRACTS } from "./toolContracts.js";
-import { TOOL_EFFECTS, annotationsFromEffects } from "./toolEffects.js";
+import {
+  TOOL_EFFECTS,
+  annotationsFromEffects,
+  type ToolEffects,
+} from "./toolEffects.js";
 
 /**
  * Descriptions that name running, launching, executing, starting, installing,
@@ -148,8 +152,19 @@ describe("tool effects", () => {
   });
 
   it("derives destructiveHint from each destructive effect alone", () => {
-    const base = TOOL_EFFECTS["binary_session"];
-    if (base === undefined) throw new Error("binary_session is unaudited");
+    // Keep annotation algebra independent of any operation's audited effects.
+    const base: ToolEffects = {
+      mutatesTarget: false,
+      mutatesSession: false,
+      writesFilesystem: false,
+      launchesProcess: false,
+      accessesNetwork: false,
+      changesUiState: false,
+      mayDiscardData: false,
+      executesForeignCode: false,
+      mutatesDevice: false,
+      idempotent: true,
+    };
     expect(annotationsFromEffects(base)).toEqual({
       readOnlyHint: true,
       destructiveHint: false,

src/contracts/toolEffects.ts

@@ -605,7 +605,7 @@ export const TOOL_EFFECTS: Readonly<Record<string, ToolEffects>> = {
     mayDiscardData: true,
     idempotent: false,
   }),
-  binary_session: effects(),
+  binary_session: effects({ launchesProcess: true }),
   export_evidence_bundle: effects({
     writesFilesystem: true,
     mayDiscardData: true,

tests/boundary/browser/cdpBrowserProvider.lifecycle_failures.test.ts

@@ -7,205 +7,12 @@ import {
 } from "../../../src/domain/browserObservation.js";
 import { analyzeWebBundleInputSchema } from "../../../src/domain/webBundleAnalysis.js";
 import { observeWebSessionInputSchema } from "../../../src/domain/browserSession.js";
-import { discoverWebMcpToolsInputSchema } from "../../../src/domain/webMcpDiscovery.js";
-import { captureWebScreenshotInputSchema } from "../../../src/domain/webScreenshot.js";
 import {
   startFakeCdpBrowser,
   type FakeCdpBrowser,
 } from "../../fixtures/fakeCdpBrowser.js";
 import { trackBrowser } from "./cdpBrowserProvider.support.js";
 
-const waitFor = async (ready: () => boolean, description: string) => {
-  const deadline = Date.now() + 5_000;
-  while (!ready()) {
-    if (Date.now() > deadline) throw new Error(description);
-    await new Promise((resolve) => setTimeout(resolve, 10));
-  }
-};
-
-describe("CdpBrowserProvider: cancellation during cleanup", () => {
-  it.each([
-    {
-      name: "WebMCP discovery",
-      operation: "discover_webmcp_tools",
-      invoke: (
-        provider: CdpBrowserProvider,
-        browser: FakeCdpBrowser,
-        signal: AbortSignal,
-      ) =>
-        provider.discoverWebMcpTools(
-          discoverWebMcpToolsInputSchema.parse({
-            cdp_endpoint: browser.endpoint,
-            allowed_origins: [browser.allowedOrigin],
-            target_id: "allowed-page",
-            observation_ms: 0,
-          }),
-          { signal },
-        ),
-    },
-    {
-      name: "screenshot capture",
-      operation: "capture_web_screenshot",
-      invoke: (
-        provider: CdpBrowserProvider,
-        browser: FakeCdpBrowser,
-        signal: AbortSignal,
-      ) =>
-        provider.captureScreenshot(
-          captureWebScreenshotInputSchema.parse({
-            cdp_endpoint: browser.endpoint,
-            allowed_origins: [browser.allowedOrigin],
-            target_id: "allowed-page",
-          }),
-          { signal },
-        ),
-    },
-  ])(
-    "reports cancellation during $name cleanup",
-    async ({ operation, invoke }) => {
-      const browser = await startFakeCdpBrowser({
-        hangOnMethod: "Page.disable",
-        webMcpTools: true,
-      });
-      trackBrowser(browser);
-      const controller = new AbortController();
-      const pending = invoke(
-        new CdpBrowserProvider(),
-        browser,
-        controller.signal,
-      );
-      await waitFor(
-        () => browser.commands.some(({ method }) => method === "Page.disable"),
-        "page cleanup did not start",
-      );
-      controller.abort();
-      await expect(pending).resolves.toMatchObject({
-        ok: false,
-        error: { _tag: "AnalysisCancelledError", operation },
-      });
-    },
-  );
-
-  it("reports cancellation when an interrupt arrives during page cleanup", async () => {
-    const browser = await startFakeCdpBrowser({
-      hangOnMethod: "Network.disable",
-    });
-    trackBrowser(browser);
-    const controller = new AbortController();
-    const pending = new CdpBrowserProvider().inspectPage(
-      inspectWebPageInputSchema.parse({
-        cdp_endpoint: browser.endpoint,
-        allowed_origins: [browser.allowedOrigin],
-        target_id: "allowed-page",
-        observation_ms: 0,
-      }),
-      { signal: controller.signal },
-    );
-    await waitFor(
-      () => browser.commands.some(({ method }) => method === "Network.disable"),
-      "page cleanup did not start",
-    );
-    controller.abort();
-
-    await expect(pending).resolves.toMatchObject({
-      ok: false,
-      error: {
-        _tag: "AnalysisCancelledError",
-        operation: "inspect_web_page",
-      },
-    });
-  });
-
-  it("reports cancellation when an interrupt arrives during session cleanup", async () => {
-    const browser = await startFakeCdpBrowser({
-      hangOnMethod: "Network.disable",
-    });
-    trackBrowser(browser);
-    const controller = new AbortController();
-    const pending = new CdpBrowserProvider().observeSession(
-      observeWebSessionInputSchema.parse({
-        cdp_endpoint: browser.endpoint,
-        allowed_origins: [browser.allowedOrigin],
-        target_id: "allowed-page",
-        observation_ms: 1,
-      }),
-      { signal: controller.signal },
-    );
-    await waitFor(
-      () => browser.commands.some(({ method }) => method === "Network.disable"),
-      "session cleanup did not start",
-    );
-    controller.abort();
-
-    await expect(pending).resolves.toMatchObject({
-      ok: false,
-      error: {
-        _tag: "AnalysisCancelledError",
-        operation: "observe_web_session",
-      },
-    });
-  });
-
-  it("reports cancellation when an interrupt arrives during source-map download", async () => {
-    const browser = await startFakeCdpBrowser({ hangSourceMap: true });
-    trackBrowser(browser);
-    const controller = new AbortController();
-    const pending = new CdpBrowserProvider().analyzeBundle(
-      analyzeWebBundleInputSchema.parse({
-        cdp_endpoint: browser.endpoint,
-        allowed_origins: [browser.allowedOrigin],
-        target_id: "allowed-page",
-        observation_ms: 0,
-        fetch_source_maps: true,
-      }),
-      { signal: controller.signal },
-    );
-    await waitFor(
-      () =>
-        browser.httpRequests.some(({ url }) => url.startsWith("/app.js.map")),
-      "source-map download did not start",
-    );
-    controller.abort();
-
-    await expect(pending).resolves.toMatchObject({
-      ok: false,
-      error: {
-        _tag: "AnalysisCancelledError",
-        operation: "analyze_web_bundle",
-      },
-    });
-  });
-
-  it("classifies custom source-map abort reasons as cancellation", async () => {
-    const browser = await startFakeCdpBrowser({ hangSourceMap: true });
-    trackBrowser(browser);
-    const controller = new AbortController();
-    const pending = new CdpBrowserProvider().analyzeBundle(
-      analyzeWebBundleInputSchema.parse({
-        cdp_endpoint: browser.endpoint,
-        allowed_origins: [browser.allowedOrigin],
-        target_id: "allowed-page",
-        observation_ms: 0,
-        fetch_source_maps: true,
-      }),
-      { signal: controller.signal },
-    );
-    await waitFor(
-      () =>
-        browser.httpRequests.some(({ url }) => url.startsWith("/app.js.map")),
-      "source-map download did not start",
-    );
-    controller.abort(new Error("stopped"));
-    await expect(pending).resolves.toMatchObject({
-      ok: false,
-      error: {
-        _tag: "AnalysisCancelledError",
-        operation: "analyze_web_bundle",
-      },
-    });
-  });
-});
-
 describe("CdpBrowserProvider: cancellation and session events", () => {
   it("honors cancellation during the final progress callback before publishing inspection", async () => {
     const browser = await startFakeCdpBrowser();

tests/boundary/browser/cdpCleanupCancellation.test.ts

@@ -73,13 +73,7 @@ it("releases cleanup when cancellation arrives after capture completes", async (
   });
   controller.abort();
   const result = await pending;
-  expect(result).toMatchObject({
-    ok: false,
-    error: {
-      _tag: "AnalysisCancelledError",
-      operation: "observe_web_session",
-    },
-  });
+  expect(result.ok).toBe(true);
   expect(browser.commands.map(({ method }) => method)).toContain(
     "Target.detachFromTarget",
   );

tests/evaluation/package/packagePlatformDiagnostics.test.ts

@@ -1,6 +1,9 @@
 import { describe, expect, it } from "vitest";
 
-import { assertLinuxPackageProviderFailure } from "../../../scripts/verify-package-platform.mjs";
+import {
+  assertLinuxPackageProviderFailure,
+  verifyPackagePlatform,
+} from "../../../scripts/verify-package-platform.mjs";
 
 const missingDisplay = {
   code: "provider_unavailable",
@@ -22,6 +25,37 @@ const execution = (failure: unknown, status = 1) => ({
   stdout: JSON.stringify(failure),
 });
 
+const overview = (procedure_count?: unknown, operation = "binary_overview") => ({
+  operation,
+  ...(procedure_count === undefined
+    ? {}
+    : { normalized_result: { procedure_count } }),
+});
+
+const verifyMacOS = async (
+  analyze: unknown,
+  inspect: unknown,
+): Promise<void> => {
+  const outputs: Readonly<Record<string, unknown>> = {
+    analyze,
+    inspect,
+    function: {
+      operation: "analyze_function",
+      provider: { id: "hopper" },
+      normalized_result: { procedure: { address: "0x1000" } },
+    },
+    xrefs: { operation: "xrefs", normalized_result: ["0x1000"] },
+    trace: { operation: "trace_feature" },
+  };
+  await verifyPackagePlatform({
+    cli: "fixture-cli",
+    environment: {},
+    platform: "darwin",
+    runCommand: async (_cli, arguments_) =>
+      JSON.stringify(outputs[arguments_[0]!]) ?? "",
+  });
+};
+
 describe("packaged Linux failure verification", () => {
   it("accepts the unsupported fixture build or an exact missing display dependency", () => {
     expect(() =>
@@ -71,3 +105,29 @@ describe("packaged Linux failure verification", () => {
     },
   );
 });
+
+describe("packaged analyze and inspect result guards", () => {
+  it("accepts positive integer procedure counts for both commands", async () => {
+    await verifyMacOS(overview(1), overview(2));
+  });
+
+  const malformed = [
+    ["missing result", undefined],
+    ["null result", null],
+    ["missing normalized count", overview()],
+    ["zero count", overview(0)],
+    ["non-integer count", overview(1.5)],
+    ["wrong operation", overview(2, "other_operation")],
+  ] as const;
+
+  it.each(["analyze", "inspect"] as const)(
+    "rejects malformed %s results before trusting the package",
+    async (command) => {
+      for (const [label, invalid] of malformed) {
+        const analyze = command === "analyze" ? invalid : overview(4);
+        const inspect = command === "inspect" ? invalid : overview(4);
+        await expect(verifyMacOS(analyze, inspect), label).rejects.toThrow();
+      }
+    },
+  );
+});

tests/fixtures/fakeCdpBrowser.ts

@@ -229,11 +229,6 @@ const serveSourceMap = (
   response: ServerResponse,
   options: FakeOptions,
 ): boolean => {
-  if (
-    options.hangSourceMap === true &&
-    request.url?.startsWith("/app.js.map") === true
-  )
-    return true;
   if (options.sourceMapBody === undefined) return false;
   if (request.url?.startsWith("/app.js.map") === true) {
     if (options.sourceMapRedirect !== undefined)

tests/fixtures/fakeCdpBrowserTypes.ts

@@ -95,8 +95,6 @@ export interface FakeOptions {
   readonly webSocketWithoutFrames?: boolean;
   readonly invalidBinaryWebSocketEvent?: boolean;
   readonly sourceMapBody?: string;
-  /** Accept a source-map request and leave the response open. */
-  readonly hangSourceMap?: boolean;
   readonly sourceMapRedirect?: string;
   readonly sessionTimeline?:
     | "same_origin"