Merge pull request #872 from corbitsdev/cl-7618-materialize-leisure-on-all-tool-results-not-an-allowlist

Spill oversized tool results without a name allowlist

Author: TheGreatAxios

docs/ARCHITECTURE.md

@@ -363,7 +363,7 @@ tool call
 
 **Rejection behavior:** Any plugin can short-circuit by returning a `ToolResult` with `isError: true`; the error propagates to the agent and downstream plugins/execution are skipped.
 
-- **Result truncation / leisure materialization** (`result-truncation-plugin.ts`, `tool-result-materialize.ts`) — Caps model-facing tool results at 10,000 chars (aligned with the reactor size-cap). Over the gate, content is leisure-materialized first (minified JSON → pretty `application/json`; NDJSON preserved; else `text/plain`), then the formatted bytes are spilled to the session blob store under `{callId}:full` and truncated inline with a `tool-output:///` URI plus absolute `contextDir/tool-output/…` path when plumbed. Under-gate results are unchanged (no pretty, no spill). MCP tools apply the same scrub-then-truncate path via `mcpClientToAgentTools` since they skip the posix middleware chain.
+- **Result truncation / leisure materialization** (`result-truncation-plugin.ts`, `tool-result-materialize.ts`) — Caps model-facing tool results at 10,000 chars (aligned with the reactor size-cap). Over the gate, any non-error result is leisure-materialized first (minified JSON → pretty `application/json`; NDJSON preserved; else `text/plain`), then the formatted bytes are spilled to the session blob store under `{callId}:full` and truncated inline with a `tool-output:///` URI plus absolute `contextDir/tool-output/…` path when plumbed. Under-gate results are unchanged (no pretty, no spill). Posix tools go through the middleware in `buildCorePosixToolPlugins` (Codex `posixTools.run` included). Fleet AgentTools (`wait_agents`, `search_agents`, …) skip that posix chain, so the same helper wraps them at mount in `createAgentToolset` and nested `runSubAgent`. MCP tools apply the same scrub-then-truncate path via `mcpClientToAgentTools` since they skip both.
 - **Path Escape** (`path-escape-plugin.ts`) — Canonicalizes path-like arguments against `cwd` and blocks `..` escapes, except into a root the permission layer's worktree-roots provider allowlists (e.g. a sibling git worktree of the same repo). Runs first so later plugins see resolved paths.
 - **Tool-output URI** (`tool-output-uri-plugin.ts`) — Normalizes mistaken `read_file` blob URIs to `tool-output:///id` (corbits-only; interchange stays unpatched).
 - **Secret Guard** (`secret-guard-plugin.ts`) — Hard-denies path-keyed tool calls (`read_file`, `write_file`, …) that would put a sensitive file into (or write it from) the model context. Runs before the permission plugin, so the path-arg deny holds even under `--dangerously-skip-permissions`. Shell commands that _reference_ a sensitive path (tokenized so `cat .env`, `bun --env-file=.env run …`, and quote/env-assignment forms are detected) are not hard-denied here: they require operator approval via the permission gate, and auto mode forces an ask through the auto-shell policy (`sensitive-path` rule). Once the operator approves, the command runs. Shell detection is best-effort: token matching defeats quoting and env-assignment/redirection forms but not dynamic path construction (variable indirection, `printf` assembly). Tool-result secret scrub still redacts credential-shaped output.

src/agent/tools.ts

@@ -21,7 +21,11 @@ import type { PermissionGate } from "../permission/gate.js";
 import { buildCorePosixToolPlugins } from "./posix-tool-plugins.js";
 import { createLazyBlobReader } from "./lazy-blob-reader.js";
 import type { BlobReader } from "@intx/types/runtime";
-import type { SpillBlobWriter } from "../plugins/result-truncation-plugin.js";
+import {
+  wrapAgentToolResultTruncation,
+  wrapAgentToolsWithResultTruncation,
+  type SpillBlobWriter,
+} from "../plugins/result-truncation-plugin.js";
 import {
   connectMCPServer as connectMCPClient,
   type MCPClient,
@@ -413,6 +417,10 @@ export async function createAgentToolset(
     );
   }
 
+  const truncationOptions = {
+    ...(getBlobWriter !== undefined ? { getBlobWriter } : {}),
+    ...(getContextDir !== undefined ? { getContextDir } : {}),
+  };
   const posixTools = createPosixTools({
     cwd,
     ...(sessionBlobReader !== undefined
@@ -426,8 +434,7 @@ export async function createAgentToolset(
       ...(sessionBlobReader !== undefined
         ? { readFileGuard: { blobReader: sessionBlobReader } }
         : {}),
-      ...(getBlobWriter !== undefined ? { getBlobWriter } : {}),
-      ...(getContextDir !== undefined ? { getContextDir } : {}),
+      ...truncationOptions,
       ...(shellEnv !== undefined ? { shellEnv } : {}),
       getBackgroundShellRegistry: () => backgroundShells,
     }),
@@ -698,8 +705,9 @@ export async function createAgentToolset(
     }),
   );
 
-  const primaryTools = baseTools.filter(
-    (tool) => tool.definition.name !== "apply_patch",
+  const primaryTools = wrapAgentToolsWithResultTruncation(
+    baseTools.filter((tool) => tool.definition.name !== "apply_patch"),
+    truncationOptions,
   );
 
   const dynamicRunner = createDynamicToolRunner(primaryTools, toolWatchdog);
@@ -796,9 +804,10 @@ export async function createAgentToolset(
   };
 
   const mountWebFetch = (tool: AgentTool): void => {
+    const wrapped = wrapAgentToolResultTruncation(tool, truncationOptions);
     dynamicRunner.removeTools(["web_fetch"]);
-    dynamicRunner.addTools([tool]);
-    replaceInheritedTool("web_fetch", tool);
+    dynamicRunner.addTools([wrapped]);
+    replaceInheritedTool("web_fetch", wrapped);
   };
 
   const swapBuiltinExaToNative = (): void => {

src/plugins/result-truncation-plugin.test.ts

@@ -11,6 +11,9 @@ import {
   resultTruncationPlugin,
   spillBlobKey,
   truncateToolResultContent,
+  wrapAgentToolResultTruncation,
+  wrapAgentToolsWithResultTruncation,
+  type SpillBlobWriter,
 } from "./result-truncation-plugin.js";
 import { toolOutputAbsolutePath } from "./tool-result-materialize.js";
 import { CREDENTIAL_REDACTION } from "./tool-result-secret-scrub.js";
@@ -373,6 +376,239 @@ describe("resultTruncationPlugin", () => {
     expect(result.content).toEqual(record);
     expect(store.blobs.size).toBe(0);
   });
+
+  test("spills oversized minified fleet JSON for wait_agents and list_agents", async () => {
+    const store = fakeBlobStore();
+    const obj = {
+      results: Array.from({ length: 80 }, (_, i) => ({
+        agent_id: `agent-${i}`,
+        report: "x".repeat(200),
+      })),
+      timed_out: false,
+    };
+    const minified = JSON.stringify(obj);
+    expect(minified.length).toBeGreaterThan(MAX_RESULT_CHARS);
+    const pretty = JSON.stringify(obj, null, 2);
+    const plugin = resultTruncationPlugin({
+      getBlobWriter: () => store.writeBlob,
+    });
+    if (plugin.middleware === undefined) throw new Error("expected middleware");
+    const middleware = plugin.middleware(async (call) => ({
+      callId: call.id,
+      content: minified,
+    }));
+
+    for (const name of ["wait_agents", "list_agents"] as const) {
+      const callId = `call-${name}`;
+      const result = await middleware(
+        { id: callId, name, arguments: {} },
+        new AbortController().signal,
+      );
+      expect(typeof result.content).toBe("string");
+      expect(String(result.content).length).toBeLessThanOrEqual(
+        MAX_RESULT_CHARS,
+      );
+      const uri = `tool-output:///${spillBlobKey(callId)}`;
+      expect(String(result.content)).toContain(uri);
+      const recovered = new TextDecoder().decode(
+        await createBlobReader(store).read(uri),
+      );
+      expect(recovered).toBe(pretty);
+    }
+  });
+
+  test("spills an oversized search_agents string payload", async () => {
+    const store = fakeBlobStore();
+    const original = `Matching agent profiles:\n\n${"body ".repeat(MAX_RESULT_CHARS)}`;
+    expect(original.length).toBeGreaterThan(MAX_RESULT_CHARS);
+    const plugin = resultTruncationPlugin({
+      getBlobWriter: () => store.writeBlob,
+    });
+    if (plugin.middleware === undefined) throw new Error("expected middleware");
+    const middleware = plugin.middleware(async (call) => ({
+      callId: call.id,
+      content: original,
+    }));
+    const result = await middleware(
+      { id: "call-search", name: "search_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(String(result.content).length).toBeLessThanOrEqual(MAX_RESULT_CHARS);
+    const uri = `tool-output:///${spillBlobKey("call-search")}`;
+    expect(String(result.content)).toContain(uri);
+    const recovered = new TextDecoder().decode(
+      await createBlobReader(store).read(uri),
+    );
+    expect(recovered).toBe(original);
+  });
+
+  test("does not truncate isError results even when over the gate", async () => {
+    const store = fakeBlobStore();
+    const original = `Error: ${"x".repeat(MAX_RESULT_CHARS + 500)}`;
+    const plugin = resultTruncationPlugin({
+      getBlobWriter: () => store.writeBlob,
+    });
+    if (plugin.middleware === undefined) throw new Error("expected middleware");
+    const middleware = plugin.middleware(async (call) => ({
+      callId: call.id,
+      content: original,
+      isError: true,
+    }));
+    const result = await middleware(
+      { id: "call-err", name: "wait_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(result.content).toBe(original);
+    expect(result.isError).toBe(true);
+    expect(store.blobs.size).toBe(0);
+  });
+});
+
+describe("wrapAgentToolResultTruncation", () => {
+  test("spills oversized wait_agents JSON from a kind:full handler", async () => {
+    const store = fakeBlobStore();
+    const payload = {
+      results: [{ report: "x".repeat(MAX_RESULT_CHARS + 500) }],
+    };
+    const minified = JSON.stringify(payload);
+    expect(minified.length).toBeGreaterThan(MAX_RESULT_CHARS);
+    const pretty = JSON.stringify(payload, null, 2);
+    const wrapped = wrapAgentToolResultTruncation(
+      {
+        kind: "full",
+        definition: {
+          name: "wait_agents",
+          description: "wait",
+          inputSchema: { type: "object" },
+        },
+        handler: async (call) => ({ callId: call.id, content: minified }),
+      },
+      { getBlobWriter: () => store.writeBlob },
+    );
+    if (wrapped.kind !== "full") throw new Error("expected full tool");
+    const result = await wrapped.handler(
+      { id: "call-wrap-wait", name: "wait_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(String(result.content).length).toBeLessThanOrEqual(MAX_RESULT_CHARS);
+    const uri = `tool-output:///${spillBlobKey("call-wrap-wait")}`;
+    expect(String(result.content)).toContain(uri);
+    const recovered = new TextDecoder().decode(
+      await createBlobReader(store).read(uri),
+    );
+    expect(recovered).toBe(pretty);
+  });
+
+  test("spills oversized search_agents string from a kind:string handler", async () => {
+    const store = fakeBlobStore();
+    const original = `Matching agent profiles:\n\n${"z".repeat(MAX_RESULT_CHARS + 500)}`;
+    const wrapped = wrapAgentToolResultTruncation(
+      {
+        kind: "string",
+        definition: {
+          name: "search_agents",
+          description: "search",
+          inputSchema: { type: "object" },
+        },
+        handler: async () => original,
+      },
+      { getBlobWriter: () => store.writeBlob },
+    );
+    if (wrapped.kind !== "full")
+      throw new Error("expected full wrapper so spill can use callId");
+    const result = await wrapped.handler(
+      { id: "call-wrap-search", name: "search_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(String(result.content).length).toBeLessThanOrEqual(MAX_RESULT_CHARS);
+    const uri = `tool-output:///${spillBlobKey("call-wrap-search")}`;
+    expect(String(result.content)).toContain(uri);
+    const recovered = new TextDecoder().decode(
+      await createBlobReader(store).read(uri),
+    );
+    expect(recovered).toBe(original);
+  });
+
+  test("does not truncate isError results from a kind:full handler", async () => {
+    const store = fakeBlobStore();
+    const original = `Error: ${"e".repeat(MAX_RESULT_CHARS + 500)}`;
+    const wrapped = wrapAgentToolResultTruncation(
+      {
+        kind: "full",
+        definition: {
+          name: "wait_agents",
+          description: "wait",
+          inputSchema: { type: "object" },
+        },
+        handler: async (call) => ({
+          callId: call.id,
+          content: original,
+          isError: true,
+        }),
+      },
+      { getBlobWriter: () => store.writeBlob },
+    );
+    if (wrapped.kind !== "full") throw new Error("expected full tool");
+    const result = await wrapped.handler(
+      { id: "call-wrap-err", name: "wait_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(result.content).toBe(original);
+    expect(result.isError).toBe(true);
+    expect(store.blobs.size).toBe(0);
+  });
+});
+
+describe("wrapAgentToolsWithResultTruncation", () => {
+  test("late-binds a blob writer after wrap so oversized wait_agents JSON spills to a readable URI", async () => {
+    const payload = {
+      results: [{ report: "n".repeat(MAX_RESULT_CHARS + 500) }],
+    };
+    const minified = JSON.stringify(payload);
+    expect(minified.length).toBeGreaterThan(MAX_RESULT_CHARS);
+    const pretty = JSON.stringify(payload, null, 2);
+
+    const childSpill: { writer?: SpillBlobWriter } = {};
+    const [wrapped] = wrapAgentToolsWithResultTruncation(
+      [
+        {
+          kind: "full",
+          definition: {
+            name: "wait_agents",
+            description: "wait",
+            inputSchema: { type: "object" },
+          },
+          handler: async (call) => ({ callId: call.id, content: minified }),
+        },
+      ],
+      { getBlobWriter: () => childSpill.writer },
+    );
+    if (wrapped === undefined || wrapped.kind !== "full") {
+      throw new Error("expected full wrapped tool");
+    }
+
+    const before = await wrapped.handler(
+      { id: "call-nested-before", name: "wait_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(String(before.content)).toContain("NOT retrievable");
+    expect(String(before.content)).not.toContain("tool-output:///");
+
+    const store = fakeBlobStore();
+    childSpill.writer = store.writeBlob;
+    const result = await wrapped.handler(
+      { id: "call-nested-wait", name: "wait_agents", arguments: {} },
+      new AbortController().signal,
+    );
+    expect(String(result.content).length).toBeLessThanOrEqual(MAX_RESULT_CHARS);
+    expect(String(result.content)).not.toContain("NOT retrievable");
+    const uri = `tool-output:///${spillBlobKey("call-nested-wait")}`;
+    expect(String(result.content)).toContain(uri);
+    const recovered = new TextDecoder().decode(
+      await createBlobReader(store).read(uri),
+    );
+    expect(recovered).toBe(pretty);
+  });
 });
 
 describe("scrub-before-spill", () => {