Fix incorrect stop reason after tool call rejection (#153 )

We were mapping tool call rejections to actual "refusals" which isn't quite right in ACP. Also filters out more of these status messages that Claude is returning.
Update to claude-agent-sdk 0.1.37 (#152 )
2026-10-27 12:40:32 +00:00 · 2025-11-13 11:19:48 +01:00 · 2025-11-13 09:56:19 +00:00 · 2025-11-13 10:42:21 +01:00 · 2025-11-05 11:36:02 +01:00 · 2025-11-04 21:34:16 -05:00
18 changed files with 2563 additions and 1658 deletions
--- a/.eslintignore
+++ b/.eslintignore
@@ -1,6 +0,0 @@
-node_modules/
-dist/
-coverage/
-*.min.js
-*.config.js
-.github/
--- a/.eslintrc.json
+++ b/.eslintrc.json
@@ -1,33 +0,0 @@
-{
-  "root": true,
-  "parser": "@typescript-eslint/parser",
-  "parserOptions": {
-    "ecmaVersion": 2020,
-    "sourceType": "module",
-    "project": "./tsconfig.json"
-  },
-  "plugins": ["@typescript-eslint"],
-  "extends": ["eslint:recommended", "plugin:@typescript-eslint/recommended", "prettier"],
-  "env": {
-    "node": true,
-    "es2020": true
-  },
-  "rules": {
-    "@typescript-eslint/explicit-function-return-type": "off",
-    "@typescript-eslint/no-explicit-any": "off",
-    "@typescript-eslint/no-unused-vars": [
-      "error",
-      {
-        "argsIgnorePattern": "^_"
-      }
-    ],
-    "no-console": "off",
-    "no-constant-condition": "off",
-    "default-case": "error",
-    "prefer-const": "error",
-    "no-var": "error",
-    "eqeqeq": ["error", "always"],
-    "curly": ["error", "all"]
-  },
-  "ignorePatterns": ["dist", "node_modules", "*.js"]
-}
--- a/.github/workflows/publish.yml
+++ b/.github/workflows/publish.yml
@@ -4,8 +4,9 @@ on:
    tags:
      - "v*.*.*"
 jobs:
-  build:
+  publish-npm:
    runs-on: ubuntu-latest
+    environment: release # Optional: for enhanced security
    permissions:
      contents: read
      id-token: write
@@ -16,8 +17,20 @@ jobs:
        with:
          node-version: "lts/*"
          registry-url: "https://registry.npmjs.org"
+      - name: Update npm
+        run: npm install -g npm@latest
      - run: npm ci
-        # Add back provenance once repo is public
-      - run: npm publish # --provenance --access public
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+      - run: npm run build
+      - run: npm publish
+  release:
+    runs-on: ubuntu-latest
+    needs: [publish-npm]
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8
+      - name: Create Release
+        id: create_release
+        uses: softprops/action-gh-release@6da8fa9354ddfdc4aeace5fc48d7f679b5214090
+        with:
+          draft: true
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,3 @@
 node_modules
 dist
+.DS_Store
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -0,0 +1,150 @@
+# Changelog
+
+## 0.10.2
+
+- Fix incorrect stop reason for tool call refusals
+
+## 0.10.1
+
+- Add additional structured metadata to tool calls
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.37
+
+## 0.10.0
+
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.30
+- Use `canUseTool` callback instead of launching an HTTP MCP server for permission checks.
+
+## 0.9.0
+
+- Support slash commands coming from MCP servers (Prompts)
+
+## 0.8.0
+
+- Revert changes to filename for cli entrypoint
+- Provide library entrypoint via lib.ts
+
+## 0.7.0
+
+- Allow importing from this package as a library in addition to running it as a CLI. Allows for easier integration into existing node applications.
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.27
+
+## 0.6.10
+
+- Provide `agentInfo` on initialization response.
+- Update to @agentclientprotocol/sdk@0.5.1
+- Fix crash when receiving a hook_response event
+- Fix for invalid locations when read call has no path
+
+## 0.6.9
+
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.26
+- Update to @agentclientprotocol/sdk@0.5.0
+
+## 0.6.8
+
+- Fix for duplicate tokens appearing in thread with streaming enabled
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.23
+- Update to @agentclientprotocol/sdk@0.4.9
+
+## 0.6.7
+
+- Fix for invalid plan input from the model introduced in latest agent-sdk
+
+## 0.6.6
+
+- Do not enable bypassPermissions mode if in root/sudo mode, because Claude Code will not start
+
+## 0.6.5
+
+- Fix for duplicated text content after streaming
+
+## 0.6.4
+
+- Support streaming partial messages!
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.21
+
+## 0.6.3
+
+- Fix issue where slash commands were loaded before initialization was complete.
+
+## 0.6.2
+
+- Fix bug where mode selection would sometimes fire before initialization was complete.
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.19
+
+## 0.6.1
+
+- Fix to allow bypassPermissions mode to be selected (it wasn't permitted previously)
+
+## 0.6.0
+
+- Provide a model selector. We use the "default" model by default, and the user can change it via the client.
+- Make sure writes require permissions when necessary: https://github.com/zed-industries/claude-code-acp/pull/92
+- Add support for appending or overriding the system prompt: https://github.com/zed-industries/claude-code-acp/pull/91
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.15
+- Update to @agentclientprotocol/sdk@0.4.8
+
+## 0.5.5
+
+- Migrate to @agentclientprotocol/sdk@0.4.5
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.13
+
+## 0.5.4
+
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.11
+- Enable setting CLAUDE_CODE_EXECUTABLE to override the executable used by the SDK https://github.com/zed-industries/claude-code-acp/pull/86
+
+## 0.5.3
+
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.8
+- Update to @zed-industries/agent-client-protocol@v0.4.5
+
+## 0.5.2
+
+- Add back @anthropic-ai/claude-code@2.0.1 as runtime dependency
+
+## 0.5.1
+
+- Update to @anthropic-ai/claude-agent-sdk@v0.1.1
+- Make improvements to ACP tools provided to the model
+
+## 0.5.0
+
+- Migrate to @anthropic-ai/claude-agent-sdk@v0.1.0
+
+## v0.4.7
+
+- More efficient file reads from the client.
+
+## v0.4.6
+
+- Update to @anthropic-ai/claude-code@v1.0.128
+
+## v0.4.5
+
+- Update to @anthropic-ai/claude-code@v1.0.124
+- Update to @zed-industries/agent-client-protocol@v0.4.3
+
+## v0.4.4
+
+- Update to @anthropic-ai/claude-code@v1.0.123
+- Update to @zed-industries/agent-client-protocol@v0.4.2
+
+## v0.4.3
+
+- Move ACP tools over MCP from an "http" MCP server to an "sdk" one so more tool calls can stay in-memory.
+- Update to @anthropic-ai/claude-code@v1.0.119
+- Update to @zed-industries/agent-client-protocol@v0.4.0
+
+## v0.4.2
+
+- Fix missing package.json metadata
+
+## v0.4.1
+
+- Add support for /compact command [ecfd36a](https://github.com/zed-industries/claude-code-acp/commit/ecfd36afa6c4e31f12e1daf9b8a2bdc12dda1794)
+- Add default limits to read tool [7bd1638](https://github.com/zed-industries/claude-code-acp/commit/7bd163818bb959b11fd2c933eff73ad83c57abb8)
+- Better rendering of Tool errors [491efe3](https://github.com/zed-industries/claude-code-acp/commit/491efe32e8547075842e448d873fc01b2ffabf3a)
+- Load managed-settings.json [f691024](https://github.com/zed-industries/claude-code-acp/commit/f691024350362858e00b97248ac68e356d2331c2)
+- Update to @anthropic-ai/claude-code@v1.0.113
+- Update to @zed-industries/agent-client-protocol@v0.3.1
--- a/README.md
+++ b/README.md
@@ -32,9 +32,13 @@ Read the docs on [External Agent](https://zed.dev/docs/ai/external-agents) suppo

 ### Other clients

- Neovim via [CodeCompanion.nvim](https://codecompanion.olimorris.dev/configuration/adapters#setup-claude-code-via-acp)
+- Emacs via [agent-shell.el](https://github.com/xenodium/agent-shell)
+- [marimo notebook](https://github.com/marimo-team/marimo)
+- Neovim
+  - via [CodeCompanion.nvim](https://codecompanion.olimorris.dev/configuration/adapters#setup-claude-code-via-acp)
+  - via [yetone/avante.nvim](https://github.com/yetone/avante.nvim)

-Setup instructions for other clients are coming soon. Feel free to [submit a PR](https://github.com/zed-industries/claude-code-acp/pulls) to add yours!
+[Submit a PR](https://github.com/zed-industries/claude-code-acp/pulls) to add yours!

 #### Installation

--- a/docs/RELEASES.md
+++ b/docs/RELEASES.md
@@ -0,0 +1,24 @@
+## How to publish a new release
+
+1. Create a new branch prep-vX.X.X
+
+```sh
+git checkout -b prep-vX.X.X
+```
+
+2. Bump the version in package.json
+
+```sh
+npm version vX.X.X
+```
+
+3. Add a new entry in CHANGELOG.md
+4. Create a PR for `prev-vX.X.X` & merge it
+5. Create tag and push it
+
+```sh
+git tag vX.X.X
+git push origin vX.X.X
+```
+
+6. Paste CHANGELOG into release notes and publish the version on GitHub
--- a/eslint.config.js
+++ b/eslint.config.js
@@ -0,0 +1,48 @@
+import js from "@eslint/js";
+import tseslint from "@typescript-eslint/eslint-plugin";
+import tsparser from "@typescript-eslint/parser";
+import prettierConfig from "eslint-config-prettier";
+import globals from "globals";
+
+export default [
+  {
+    ignores: ["node_modules/", "dist/", "coverage/", "*.min.js", "*.config.js", ".github/"],
+  },
+  js.configs.recommended,
+  {
+    files: ["src/**/*.ts"],
+    languageOptions: {
+      ecmaVersion: 2020,
+      sourceType: "module",
+      parser: tsparser,
+      parserOptions: {
+        project: "./tsconfig.json",
+      },
+      globals: {
+        ...globals.node,
+      },
+    },
+    plugins: {
+      "@typescript-eslint": tseslint,
+    },
+    rules: {
+      ...tseslint.configs.recommended.rules,
+      "@typescript-eslint/explicit-function-return-type": "off",
+      "@typescript-eslint/no-explicit-any": "off",
+      "@typescript-eslint/no-unused-vars": [
+        "error",
+        {
+          argsIgnorePattern: "^_",
+        },
+      ],
+      "no-console": "off",
+      "no-constant-condition": "off",
+      "default-case": "error",
+      "prefer-const": "error",
+      "no-var": "error",
+      eqeqeq: ["error", "always"],
+      curly: ["error", "all"],
+    },
+  },
+  prettierConfig,
+];
--- a/package-lock.json
+++ b/package-lock.json
--- a/package.json
+++ b/package.json
@@ -3,12 +3,12 @@
  "publishConfig": {
    "access": "public"
  },
+  "version": "0.10.2",
+  "description": "An ACP-compatible coding agent powered by the Claude Code SDK (TypeScript)",
+  "main": "dist/lib.js",
  "bin": {
    "claude-code-acp": "./dist/index.js"
  },
-  "version": "0.2.5",
-  "description": "An ACP-compatible coding agent powered by the Claude Code SDK (TypeScript)",
-  "main": "dist/index.js",
  "type": "module",
  "files": [
    "dist/",
@@ -40,30 +40,37 @@
    "sdk",
    "code"
  ],
+  "homepage": "https://github.com/zed-industries/claude-code-acp#readme",
+  "bugs": {
+    "url": "https://github.com/zed-industries/claude-code-acp/issues"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/zed-industries/claude-code-acp.git"
+  },
  "author": "Zed Industries",
  "license": "Apache-2.0",
  "dependencies": {
-    "@anthropic-ai/claude-code": "^1.0.100",
-    "@modelcontextprotocol/sdk": "^1.17.4",
-    "@types/express": "^5.0.3",
-    "@zed-industries/agent-client-protocol": "0.2.0-alpha.6",
-    "diff": "^8.0.2",
-    "express": "^5.1.0",
-    "minimist": "^1.2.8",
-    "uuid": "11.1.0"
+    "@agentclientprotocol/sdk": "0.5.1",
+    "@anthropic-ai/claude-agent-sdk": "0.1.37",
+    "@anthropic-ai/claude-code": "2.0.37",
+    "@modelcontextprotocol/sdk": "1.21.1",
+    "diff": "8.0.2",
+    "uuid": "13.0.0"
  },
  "devDependencies": {
-    "@types/minimist": "^1.2.5",
-    "@types/node": "^24.3.0",
-    "@types/uuid": "^10.0.0",
-    "@typescript-eslint/eslint-plugin": "^8.42.0",
-    "@typescript-eslint/parser": "^8.42.0",
-    "@vitest/ui": "^3.2.4",
-    "eslint": "^8.57.0",
-    "eslint-config-prettier": "^9.1.0",
-    "prettier": "^3.6.2",
-    "ts-node": "^10.9.2",
-    "typescript": "^5.4.0",
-    "vitest": "^3.2.4"
+    "@anthropic-ai/sdk": "0.68.0",
+    "@types/node": "24.10.1",
+    "@types/uuid": "10.0.0",
+    "@typescript-eslint/eslint-plugin": "8.46.4",
+    "@typescript-eslint/parser": "8.46.4",
+    "@vitest/ui": "4.0.8",
+    "eslint": "9.39.1",
+    "eslint-config-prettier": "10.1.8",
+    "globals": "16.5.0",
+    "prettier": "3.6.2",
+    "ts-node": "10.9.2",
+    "typescript": "5.9.3",
+    "vitest": "4.0.8"
  }
 }
--- a/src/acp-agent.ts
+++ b/src/acp-agent.ts
@@ -7,6 +7,7 @@ import {
  ClientCapabilities,
  InitializeRequest,
  InitializeResponse,
+  ndJsonStream,
  NewSessionRequest,
  NewSessionResponse,
  PromptRequest,
@@ -14,33 +15,50 @@ import {
  ReadTextFileRequest,
  ReadTextFileResponse,
  RequestError,
+  SessionModelState,
+  SetSessionModelRequest,
+  SetSessionModelResponse,
+  SetSessionModeRequest,
+  SetSessionModeResponse,
  TerminalHandle,
  TerminalOutputResponse,
  WriteTextFileRequest,
  WriteTextFileResponse,
-} from "@zed-industries/agent-client-protocol";
+} from "@agentclientprotocol/sdk";
 import {
+  CanUseTool,
  McpServerConfig,
  Options,
+  PermissionMode,
  Query,
  query,
-  SDKAssistantMessage,
+  SDKPartialAssistantMessage,
  SDKUserMessage,
-} from "@anthropic-ai/claude-code";
+} from "@anthropic-ai/claude-agent-sdk";
 import * as fs from "node:fs";
 import * as path from "node:path";
 import * as os from "node:os";
 import { v7 as uuidv7 } from "uuid";
 import { nodeToWebReadable, nodeToWebWritable, Pushable, unreachable } from "./utils.js";
-import { SessionNotification } from "@zed-industries/agent-client-protocol";
-import { createMcpServer } from "./mcp-server.js";
-import { AddressInfo } from "node:net";
-import { toolInfoFromToolUse, planEntries, toolUpdateFromToolResult } from "./tools.js";
+import { SessionNotification } from "@agentclientprotocol/sdk";
+import { createMcpServer, EDIT_TOOL_NAMES, toolNames } from "./mcp-server.js";
+import {
+  toolInfoFromToolUse,
+  planEntries,
+  toolUpdateFromToolResult,
+  ClaudePlanEntry,
+  registerHookCallback,
+  postToolUseHook,
+} from "./tools.js";
+import { ContentBlockParam } from "@anthropic-ai/sdk/resources";
+import { BetaContentBlock, BetaRawContentBlockDelta } from "@anthropic-ai/sdk/resources/beta.mjs";
+import packageJson from "../package.json" with { type: "json" };

 type Session = {
  query: Query;
  input: Pushable<SDKUserMessage>;
  cancelled: boolean;
+  permissionMode: PermissionMode;
 };

 type BackgroundTerminal =
@@ -54,10 +72,30 @@ type BackgroundTerminal =
      pendingOutput: TerminalOutputResponse;
    };

-type ToolUseCache = {
-  [key: string]: { type: "tool_use"; id: string; name: string; input: any };
+/**
+ * Extra metadata that the agent provides for each tool_call / tool_update update, under the `_meta.claudeCode` key.
+ */
+export type ToolUpdateMeta = {
+  /* The name of the tool that was used in Claude Code. */
+  toolName: string;
+  /* The structured output provided by Claude Code. */
+  toolResponse?: unknown;
 };

+type ToolUseCache = {
+  [key: string]: {
+    type: "tool_use" | "server_tool_use" | "mcp_tool_use";
+    id: string;
+    name: string;
+    input: any;
+  };
+};
+
+const DEFAULT_MODEL_ID = "default";
+
+// Bypass Permissions doesn't work if we are a root/sudo user
+const IS_ROOT = (process.geteuid?.() ?? process.getuid?.()) === 0;
+
 // Implement the ACP Agent interface
 export class ClaudeAcpAgent implements Agent {
  sessions: {
@@ -78,27 +116,47 @@ export class ClaudeAcpAgent implements Agent {

  async initialize(request: InitializeRequest): Promise<InitializeResponse> {
    this.clientCapabilities = request.clientCapabilities;
+
+    // Default authMethod
+    const authMethod: any = {
+      description: "Run `claude /login` in the terminal",
+      name: "Log in with Claude Code",
+      id: "claude-login",
+    };
+
+    // If client supports terminal-auth capability, use that instead.
+    if (request.clientCapabilities?._meta?.["terminal-auth"] === true) {
+      const cliPath = import.meta
+        .resolve("@anthropic-ai/claude-agent-sdk/cli.js")
+        .replace("file://", "");
+
+      authMethod._meta = {
+        "terminal-auth": {
+          command: "node",
+          args: [cliPath, "/login"],
+          label: "Claude Code Login",
+        },
+      };
+    }
+
    return {
      protocolVersion: 1,
-      // todo!()
      agentCapabilities: {
        promptCapabilities: {
          image: true,
          embeddedContext: true,
        },
+        mcpCapabilities: {
+          http: true,
+          sse: true,
+        },
      },
-      authMethods: [
-        {
-          description: "Run `claude /login` in the terminal",
-          name: "Log in with Claude Code",
-          id: "claude-login",
-        },
-        {
-          description: "Anthropic API KEY",
-          name: "Use Anthropic API Key",
-          id: "anthropic-api-key",
-        },
-      ],
+      agentInfo: {
+        name: packageJson.name,
+        title: "Claude Code",
+        version: packageJson.version,
+      },
+      authMethods: [authMethod],
    };
  }
  async newSession(params: NewSessionRequest): Promise<NewSessionResponse> {
@@ -115,50 +173,89 @@ export class ClaudeAcpAgent implements Agent {
    const mcpServers: Record<string, McpServerConfig> = {};
    if (Array.isArray(params.mcpServers)) {
      for (const server of params.mcpServers) {
-        mcpServers[server.name] = {
-          type: "stdio",
-          command: server.command,
-          args: server.args,
-          env: server.env
-            ? Object.fromEntries(server.env.map((e) => [e.name, e.value]))
-            : undefined,
-        };
+        if ("type" in server) {
+          mcpServers[server.name] = {
+            type: server.type,
+            url: server.url,
+            headers: server.headers
+              ? Object.fromEntries(server.headers.map((e) => [e.name, e.value]))
+              : undefined,
+          };
+        } else {
+          mcpServers[server.name] = {
+            type: "stdio",
+            command: server.command,
+            args: server.args,
+            env: server.env
+              ? Object.fromEntries(server.env.map((e) => [e.name, e.value]))
+              : undefined,
+          };
+        }
      }
    }

-    const server = await createMcpServer(this, sessionId, this.clientCapabilities);
-    const address = server.address() as AddressInfo;
+    const server = createMcpServer(this, sessionId, this.clientCapabilities);
    mcpServers["acp"] = {
-      type: "http",
-      url: "http://127.0.0.1:" + address.port + "/mcp",
-      headers: {
-        "x-acp-proxy-session-id": sessionId,
-      },
+      type: "sdk",
+      name: "acp",
+      instance: server,
    };

+    let systemPrompt: Options["systemPrompt"] = { type: "preset", preset: "claude_code" };
+    if (params._meta?.systemPrompt) {
+      const customPrompt = params._meta.systemPrompt;
+      if (typeof customPrompt === "string") {
+        systemPrompt = customPrompt;
+      } else if (
+        typeof customPrompt === "object" &&
+        "append" in customPrompt &&
+        typeof customPrompt.append === "string"
+      ) {
+        systemPrompt.append = customPrompt.append;
+      }
+    }
+
+    const permissionMode = "default";
+
    const options: Options = {
      cwd: params.cwd,
+      includePartialMessages: true,
      mcpServers,
-      permissionPromptToolName: "mcp__acp__permission",
+      systemPrompt,
+      settingSources: ["user", "project", "local"],
+      // If we want bypassPermissions to be an option, we have to allow it here.
+      // But it doesn't work in root mode, so we only activate it if it will work.
+      allowDangerouslySkipPermissions: !IS_ROOT,
+      permissionMode,
+      canUseTool: this.canUseTool(sessionId),
      stderr: (err) => console.error(err),
      // note: although not documented by the types, passing an absolute path
      // here works to find zed's managed node version.
      executable: process.execPath as any,
+      ...(process.env.CLAUDE_CODE_EXECUTABLE && {
+        pathToClaudeCodeExecutable: process.env.CLAUDE_CODE_EXECUTABLE,
+      }),
+      hooks: {
+        PostToolUse: [
+          {
+            hooks: [postToolUseHook],
+          },
+        ],
+      },
    };

    const allowedTools = [];
    const disallowedTools = [];
    if (this.clientCapabilities?.fs?.readTextFile) {
-      allowedTools.push("mcp__acp__read");
+      allowedTools.push(toolNames.read);
      disallowedTools.push("Read");
    }
    if (this.clientCapabilities?.fs?.writeTextFile) {
-      allowedTools.push("mcp__acp__write");
-      disallowedTools.push("Write", "Edit", "MultiEdit");
+      disallowedTools.push("Write", "Edit");
    }
    if (this.clientCapabilities?.terminal) {
-      allowedTools.push("mcp__acp__BashOutput", "mcp__acp__KillBash");
-      disallowedTools.push("Bash", "BashOutput", "KillBash");
+      allowedTools.push(toolNames.bashOutput, toolNames.killShell);
+      disallowedTools.push("Bash", "BashOutput", "KillShell");
    }

    if (allowedTools.length > 0) {
@@ -172,13 +269,19 @@ export class ClaudeAcpAgent implements Agent {
      prompt: input,
      options,
    });
+
    this.sessions[sessionId] = {
      query: q,
      input: input,
      cancelled: false,
+      permissionMode,
    };

-    getAvailableSlashCommands(q).then((availableCommands) => {
+    const availableCommands = await getAvailableSlashCommands(q);
+    const models = await getAvailableModels(q);
+
+    // Needs to happen after we return the session
+    setTimeout(() => {
      this.client.sessionUpdate({
        sessionId,
        update: {
@@ -186,10 +289,41 @@ export class ClaudeAcpAgent implements Agent {
          availableCommands,
        },
      });
-    });
+    }, 0);
+
+    const availableModes = [
+      {
+        id: "default",
+        name: "Always Ask",
+        description: "Prompts for permission on first use of each tool",
+      },
+      {
+        id: "acceptEdits",
+        name: "Accept Edits",
+        description: "Automatically accepts file edit permissions for the session",
+      },
+      {
+        id: "plan",
+        name: "Plan Mode",
+        description: "Claude can analyze but not modify files or execute commands",
+      },
+    ];
+    // Only works in non-root mode
+    if (!IS_ROOT) {
+      availableModes.push({
+        id: "bypassPermissions",
+        name: "Bypass Permissions",
+        description: "Skips all permission prompts",
+      });
+    }

    return {
      sessionId,
+      models,
+      modes: {
+        currentModeId: permissionMode,
+        availableModes,
+      },
    };
  }

@@ -217,6 +351,17 @@ export class ClaudeAcpAgent implements Agent {
      }
      switch (message.type) {
        case "system":
+          switch (message.subtype) {
+            case "init":
+              break;
+            case "compact_boundary":
+            case "hook_response":
+            case "status":
+              // Todo: process via status api: https://docs.claude.com/en/docs/claude-code/hooks#hook-output
+              break;
+            default:
+              unreachable(message);
+          }
          break;
        case "result": {
          if (this.sessions[params.sessionId].cancelled) {
@@ -232,36 +377,94 @@ export class ClaudeAcpAgent implements Agent {
              return { stopReason: "end_turn" };
            }
            case "error_during_execution":
-              return { stopReason: "refusal" };
+              return { stopReason: "end_turn" };
+            case "error_max_budget_usd":
            case "error_max_turns":
              return { stopReason: "max_turn_requests" };
            default:
-              return { stopReason: "refusal" };
+              unreachable(message);
          }
+          break;
        }
-        case "user":
-        case "assistant": {
-          if (this.sessions[params.sessionId].cancelled) {
-            continue;
-          }
-
-          if (
-            message.message.model === "<synthetic>" &&
-            message.message.content.length === 1 &&
-            message.message.content[0].text.includes("Please run /login")
-          ) {
-            throw RequestError.authRequired();
-          }
-          for (const notification of toAcpNotifications(
+        case "stream_event": {
+          for (const notification of streamEventToAcpNotifications(
            message,
            params.sessionId,
            this.toolUseCache,
            this.fileContentCache,
+            this.client,
          )) {
            await this.client.sessionUpdate(notification);
          }
          break;
        }
+        case "user":
+        case "assistant": {
+          if (this.sessions[params.sessionId].cancelled) {
+            break;
+          }
+
+          // Slash commands like /compact can generate invalid output... doesn't match
+          // their own docs: https://docs.anthropic.com/en/docs/claude-code/sdk/sdk-slash-commands#%2Fcompact-compact-conversation-history
+          if (
+            typeof message.message.content === "string" &&
+            message.message.content.includes("<local-command-stdout>")
+          ) {
+            console.log(message.message.content);
+            break;
+          }
+
+          if (
+            typeof message.message.content === "string" &&
+            message.message.content.includes("<local-command-stderr>")
+          ) {
+            console.error(message.message.content);
+            break;
+          }
+          // Skip these user messages for now, since they seem to just be messages we don't want in the feed
+          if (
+            message.type === "user" &&
+            (typeof message.message.content === "string" ||
+              (Array.isArray(message.message.content) &&
+                message.message.content.length === 1 &&
+                message.message.content[0].type === "text"))
+          ) {
+            break;
+          }
+
+          if (
+            message.type === "assistant" &&
+            message.message.model === "<synthetic>" &&
+            Array.isArray(message.message.content) &&
+            message.message.content.length === 1 &&
+            message.message.content[0].type === "text" &&
+            message.message.content[0].text.includes("Please run /login")
+          ) {
+            throw RequestError.authRequired();
+          }
+
+          const content =
+            message.type === "assistant"
+              ? // Handled by stream events above
+                message.message.content.filter((item) => !["text", "thinking"].includes(item.type))
+              : message.message.content;
+
+          for (const notification of toAcpNotifications(
+            content,
+            message.message.role,
+            params.sessionId,
+            this.toolUseCache,
+            this.fileContentCache,
+            this.client,
+          )) {
+            await this.client.sessionUpdate(notification);
+          }
+          break;
+        }
+        case "tool_progress":
+          break;
+        case "auth_status":
+          break;
        default:
          unreachable(message);
      }
@@ -277,6 +480,38 @@ export class ClaudeAcpAgent implements Agent {
    await this.sessions[params.sessionId].query.interrupt();
  }

+  async setSessionModel(params: SetSessionModelRequest): Promise<SetSessionModelResponse | void> {
+    if (!this.sessions[params.sessionId]) {
+      throw new Error("Session not found");
+    }
+    await this.sessions[params.sessionId].query.setModel(params.modelId);
+  }
+
+  async setSessionMode(params: SetSessionModeRequest): Promise<SetSessionModeResponse> {
+    if (!this.sessions[params.sessionId]) {
+      throw new Error("Session not found");
+    }
+
+    switch (params.modeId) {
+      case "default":
+      case "acceptEdits":
+      case "bypassPermissions":
+      case "plan":
+        this.sessions[params.sessionId].permissionMode = params.modeId;
+        try {
+          await this.sessions[params.sessionId].query.setPermissionMode(params.modeId);
+        } catch (error) {
+          const errorMessage =
+            error instanceof Error && error.message ? error.message : "Invalid Mode";
+
+          throw new Error(errorMessage);
+        }
+        return {};
+      default:
+        throw new Error("Invalid Mode");
+    }
+  }
+
  async readTextFile(params: ReadTextFileRequest): Promise<ReadTextFileResponse> {
    const response = await this.client.readTextFile(params);
    if (!params.limit && !params.line) {
@@ -290,60 +525,174 @@ export class ClaudeAcpAgent implements Agent {
    this.fileContentCache[params.path] = params.content;
    return response;
  }
+
+  canUseTool(sessionId: string): CanUseTool {
+    return async (toolName, toolInput, { suggestions, toolUseID }) => {
+      const session = this.sessions[sessionId];
+      if (!session) {
+        return {
+          behavior: "deny",
+          message: "Session not found",
+          interrupt: true,
+        };
+      }
+
+      if (toolName === "ExitPlanMode") {
+        const response = await this.client.requestPermission({
+          options: [
+            {
+              kind: "allow_always",
+              name: "Yes, and auto-accept edits",
+              optionId: "acceptEdits",
+            },
+            { kind: "allow_once", name: "Yes, and manually approve edits", optionId: "default" },
+            { kind: "reject_once", name: "No, keep planning", optionId: "plan" },
+          ],
+          sessionId,
+          toolCall: {
+            toolCallId: toolUseID,
+            rawInput: toolInput,
+          },
+        });
+
+        if (
+          response.outcome?.outcome === "selected" &&
+          (response.outcome.optionId === "default" || response.outcome.optionId === "acceptEdits")
+        ) {
+          session.permissionMode = response.outcome.optionId;
+          await this.client.sessionUpdate({
+            sessionId,
+            update: {
+              sessionUpdate: "current_mode_update",
+              currentModeId: response.outcome.optionId,
+            },
+          });
+
+          return {
+            behavior: "allow",
+            updatedInput: toolInput,
+            updatedPermissions: suggestions ?? [
+              { type: "setMode", mode: response.outcome.optionId, destination: "session" },
+            ],
+          };
+        } else {
+          return {
+            behavior: "deny",
+            message: "User rejected request to exit plan mode.",
+            interrupt: true,
+          };
+        }
+      }
+
+      if (
+        session.permissionMode === "bypassPermissions" ||
+        (session.permissionMode === "acceptEdits" && EDIT_TOOL_NAMES.includes(toolName))
+      ) {
+        return {
+          behavior: "allow",
+          updatedInput: toolInput,
+          updatedPermissions: suggestions ?? [
+            { type: "addRules", rules: [{ toolName }], behavior: "allow", destination: "session" },
+          ],
+        };
+      }
+
+      const response = await this.client.requestPermission({
+        options: [
+          {
+            kind: "allow_always",
+            name: "Always Allow",
+            optionId: "allow_always",
+          },
+          { kind: "allow_once", name: "Allow", optionId: "allow" },
+          { kind: "reject_once", name: "Reject", optionId: "reject" },
+        ],
+        sessionId,
+        toolCall: {
+          toolCallId: toolUseID,
+          rawInput: toolInput,
+        },
+      });
+      if (
+        response.outcome?.outcome === "selected" &&
+        (response.outcome.optionId === "allow" || response.outcome.optionId === "allow_always")
+      ) {
+        // If Claude Code has suggestions, it will update their settings already
+        if (response.outcome.optionId === "allow_always") {
+          return {
+            behavior: "allow",
+            updatedInput: toolInput,
+            updatedPermissions: suggestions ?? [
+              {
+                type: "addRules",
+                rules: [{ toolName }],
+                behavior: "allow",
+                destination: "session",
+              },
+            ],
+          };
+        }
+        return {
+          behavior: "allow",
+          updatedInput: toolInput,
+        };
+      } else {
+        return {
+          behavior: "deny",
+          message: "User refused permission to run tool",
+          interrupt: true,
+        };
+      }
+    };
+  }
+}
+
+async function getAvailableModels(query: Query): Promise<SessionModelState> {
+  const models = await query.supportedModels();
+
+  // Query doesn't give us access to the currently selected model, so we default to "default" if it is there,
+  // otherwise we just choose the first model in the list.
+  const currentModel = models.find((model) => model.value === DEFAULT_MODEL_ID) || models[0];
+  await query.setModel(currentModel.value);
+
+  const availableModels = models.map((model) => ({
+    modelId: model.value,
+    name: model.displayName,
+    description: model.description,
+  }));
+
+  return {
+    availableModels,
+    currentModelId: currentModel.value,
+  };
 }

 async function getAvailableSlashCommands(query: Query): Promise<AvailableCommand[]> {
  const UNSUPPORTED_COMMANDS = [
-    "add-dir",
-    "agents", // Modal
-    "bashes", // Modal
-    "bug", // Modal
-    "clear", // Escape Codes
-    "compact", // Not supported via SDK?
-    "config", // Modal
-    "context", // Escape Codes
-    "cost", // Escape Codes
-    "doctor", // Escape Codes
-    "exit",
-    "export", // Modal
-    "help", // Modal
-    "hooks", // Modal
-    "ide", // Modal
-    "install-github-app", // Modal
+    "context",
+    "cost",
    "login",
    "logout",
-    "memory",
-    "mcp",
-    "migrate-installer", // Modal
-    "model", // Not supported via SDK?
-    "output-style", // Modal
-    "output-style:new", // Modal
-    "permissions", // Modal
-    "privacy-settings",
-    "release-notes", // Escape Codes
-    "resume",
-    "status", // Not supported via SDK?
-    "statusline", // Not needed
-    "terminal-setup", // Not needed
-    "todos", // Escape Codes
-    "vim", // Not needed
+    "output-style:new",
+    "release-notes",
+    "todos",
  ];
-  //todo: Do not use `as any` once `supportedCommands` is exposed via the typescript interface
-  const commands = await (query as any).supportedCommands();
+  const commands = await query.supportedCommands();

  return commands
-    .map((command: { name: string; description: string; argumentHint: string }) => {
+    .map((command) => {
      const input = command.argumentHint ? { hint: command.argumentHint } : null;
+      let name = command.name;
+      if (command.name.endsWith(" (MCP)")) {
+        name = `mcp:${name.replace(" (MCP)", "")}`;
+      }
      return {
-        name: command.name,
+        name,
        description: command.description || "",
        input,
      };
    })
-    .filter(
-      (command: AvailableCommand) =>
-        !(command.name.match(/\(MCP\)/) || UNSUPPORTED_COMMANDS.includes(command.name)),
-    );
+    .filter((command: AvailableCommand) => !UNSUPPORTED_COMMANDS.includes(command.name));
 }

 function formatUriAsLink(uri: string): string {
@@ -363,15 +712,23 @@ function formatUriAsLink(uri: string): string {
  }
 }

-function promptToClaude(prompt: PromptRequest): SDKUserMessage {
+export function promptToClaude(prompt: PromptRequest): SDKUserMessage {
  const content: any[] = [];
  const context: any[] = [];

  for (const chunk of prompt.prompt) {
    switch (chunk.type) {
-      case "text":
-        content.push({ type: "text", text: chunk.text });
+      case "text": {
+        let text = chunk.text;
+        // change /mcp:server:command args -> /server:command (MCP) args
+        const mcpMatch = text.match(/^\/mcp:([^:\s]+):(\S+)(\s+.*)?$/);
+        if (mcpMatch) {
+          const [, server, command, args] = mcpMatch;
+          text = `/${server}:${command} (MCP)${args || ""}`;
+        }
+        content.push({ type: "text", text });
        break;
+      }
      case "resource_link": {
        const formattedUri = formatUriAsLink(chunk.uri);
        content.push({
@@ -439,20 +796,37 @@ function promptToClaude(prompt: PromptRequest): SDKUserMessage {
 * Only handles text, image, and thinking chunks for now.
 */
 export function toAcpNotifications(
-  message: SDKAssistantMessage | SDKUserMessage,
+  content: string | ContentBlockParam[] | BetaContentBlock[] | BetaRawContentBlockDelta[],
+  role: "assistant" | "user",
  sessionId: string,
  toolUseCache: ToolUseCache,
  fileContentCache: { [key: string]: string },
+  client: AgentSideConnection,
 ): SessionNotification[] {
-  const chunks = message.message.content as ContentChunk[];
+  if (typeof content === "string") {
+    return [
+      {
+        sessionId,
+        update: {
+          sessionUpdate: role === "assistant" ? "agent_message_chunk" : "user_message_chunk",
+          content: {
+            type: "text",
+            text: content,
+          },
+        },
+      },
+    ];
+  }
+
  const output = [];
  // Only handle the first chunk for streaming; extend as needed for batching
-  for (const chunk of chunks) {
+  for (const chunk of content) {
    let update: SessionNotification["update"] | null = null;
    switch (chunk.type) {
      case "text":
+      case "text_delta":
        update = {
-          sessionUpdate: "agent_message_chunk",
+          sessionUpdate: role === "assistant" ? "agent_message_chunk" : "user_message_chunk",
          content: {
            type: "text",
            text: chunk.text,
@@ -461,7 +835,7 @@ export function toAcpNotifications(
        break;
      case "image":
        update = {
-          sessionUpdate: "agent_message_chunk",
+          sessionUpdate: role === "assistant" ? "agent_message_chunk" : "user_message_chunk",
          content: {
            type: "image",
            data: chunk.source.type === "base64" ? chunk.source.data : "",
@@ -471,6 +845,7 @@ export function toAcpNotifications(
        };
        break;
      case "thinking":
+      case "thinking_delta":
        update = {
          sessionUpdate: "agent_thought_chunk",
          content: {
@@ -480,24 +855,74 @@ export function toAcpNotifications(
        };
        break;
      case "tool_use":
+      case "server_tool_use":
+      case "mcp_tool_use": {
        toolUseCache[chunk.id] = chunk;
        if (chunk.name === "TodoWrite") {
-          update = {
-            sessionUpdate: "plan",
-            entries: planEntries(chunk.input),
-          };
+          // @ts-expect-error - sometimes input is empty object
+          if (Array.isArray(chunk.input.todos)) {
+            update = {
+              sessionUpdate: "plan",
+              entries: planEntries(chunk.input as { todos: ClaudePlanEntry[] }),
+            };
+          }
        } else {
+          // Register hook callback to receive the structured output from the hook
+          registerHookCallback(chunk.id, {
+            onPostToolUseHook: async (toolUseId, toolInput, toolResponse) => {
+              const toolUse = toolUseCache[toolUseId];
+              if (toolUse) {
+                const update: SessionNotification["update"] = {
+                  _meta: {
+                    claudeCode: {
+                      toolResponse,
+                      toolName: toolUse.name,
+                    } satisfies ToolUpdateMeta,
+                  },
+                  toolCallId: toolUseId,
+                  sessionUpdate: "tool_call_update",
+                };
+                await client.sessionUpdate({
+                  sessionId,
+                  update,
+                });
+              } else {
+                console.error(
+                  `[claude-code-acp] Got a tool response for tool use that wasn't tracked: ${toolUseId}`,
+                );
+              }
+            },
+          });
+
+          let rawInput;
+          try {
+            rawInput = JSON.parse(JSON.stringify(chunk.input));
+          } catch {
+            // ignore if we can't turn it to JSON
+          }
          update = {
+            _meta: {
+              claudeCode: {
+                toolName: chunk.name,
+              } satisfies ToolUpdateMeta,
+            },
            toolCallId: chunk.id,
            sessionUpdate: "tool_call",
-            rawInput: chunk.input,
+            rawInput,
            status: "pending",
            ...toolInfoFromToolUse(chunk, fileContentCache),
          };
        }
        break;
+      }

-      case "tool_result": {
+      case "tool_result":
+      case "web_fetch_tool_result":
+      case "web_search_tool_result":
+      case "code_execution_tool_result":
+      case "bash_code_execution_tool_result":
+      case "text_editor_code_execution_tool_result":
+      case "mcp_tool_result": {
        const toolUse = toolUseCache[chunk.tool_use_id];
        if (!toolUse) {
          console.error(
@@ -508,17 +933,31 @@ export function toAcpNotifications(

        if (toolUse.name !== "TodoWrite") {
          update = {
+            _meta: {
+              claudeCode: {
+                toolName: toolUse.name,
+              } satisfies ToolUpdateMeta,
+            },
            toolCallId: chunk.tool_use_id,
            sessionUpdate: "tool_call_update",
-            status: chunk.is_error ? "failed" : "completed",
+            status: "is_error" in chunk && chunk.is_error ? "failed" : "completed",
            ...toolUpdateFromToolResult(chunk, toolUseCache[chunk.tool_use_id]),
          };
        }
        break;
      }

+      case "document":
+      case "search_result":
+      case "redacted_thinking":
+      case "input_json_delta":
+      case "citations_delta":
+      case "signature_delta":
+      case "container_upload":
+        break;
+
      default:
-        throw new Error("unhandled chunk type: " + chunk.type);
+        unreachable(chunk);
    }
    if (update) {
      output.push({ sessionId, update });
@@ -528,31 +967,49 @@ export function toAcpNotifications(
  return output;
 }

-export function runAcp() {
-  new AgentSideConnection(
-    (client) => new ClaudeAcpAgent(client),
-    nodeToWebWritable(process.stdout),
-    nodeToWebReadable(process.stdin),
-  );
+export function streamEventToAcpNotifications(
+  message: SDKPartialAssistantMessage,
+  sessionId: string,
+  toolUseCache: ToolUseCache,
+  fileContentCache: { [key: string]: string },
+  client: AgentSideConnection,
+): SessionNotification[] {
+  const event = message.event;
+  switch (event.type) {
+    case "content_block_start":
+      return toAcpNotifications(
+        [event.content_block],
+        "assistant",
+        sessionId,
+        toolUseCache,
+        fileContentCache,
+        client,
+      );
+    case "content_block_delta":
+      return toAcpNotifications(
+        [event.delta],
+        "assistant",
+        sessionId,
+        toolUseCache,
+        fileContentCache,
+        client,
+      );
+    // No content
+    case "message_start":
+    case "message_delta":
+    case "message_stop":
+    case "content_block_stop":
+      return [];
+
+    default:
+      unreachable(event);
+  }
 }

-type ContentChunk =
-  | { type: "text"; text: string }
-  | { type: "tool_use"; id: string; name: string; input: any }
-  | {
-      type: "tool_result";
-      content: string;
-      tool_use_id: string;
-      is_error: boolean;
-    } // content type depends on your Content definition
-  | { type: "thinking"; thinking: string }
-  | { type: "redacted_thinking" }
-  | { type: "image"; source: ImageSource }
-  | { type: "document" }
-  | { type: "web_search_tool_result" }
-  | { type: "untagged_text"; text: string };
+export function runAcp() {
+  const input = nodeToWebWritable(process.stdout);
+  const output = nodeToWebReadable(process.stdin);

-// Example ImageSource type (adjust as needed)
-type ImageSource =
-  | { type: "base64"; data: string; media_type: string }
-  | { type: "url"; url: string };
+  const stream = ndJsonStream(input, output);
+  new AgentSideConnection((client) => new ClaudeAcpAgent(client), stream);
+}
--- a/src/index.ts
+++ b/src/index.ts
@@ -1,5 +1,13 @@
 #!/usr/bin/env node

+// Load managed settings and apply environment variables
+import { loadManagedSettings, applyEnvironmentSettings } from "./utils.js";
+
+const managedSettings = loadManagedSettings();
+if (managedSettings) {
+  applyEnvironmentSettings(managedSettings);
+}
+
 // stdout is used to send messages to the client
 // we redirect everything else to stderr to make sure it doesn't interfere with ACP
 console.log = console.error;
@@ -11,7 +19,7 @@ process.on("unhandledRejection", (reason, promise) => {
  console.error("Unhandled Rejection at:", promise, "reason:", reason);
 });

-import { runAcp as runAcp } from "./acp-agent.js";
+import { runAcp } from "./acp-agent.js";
 runAcp();

 // Keep process alive
--- a/src/lib.ts
+++ b/src/lib.ts
@@ -0,0 +1,21 @@
+// Export the main agent class and utilities for library usage
+export {
+  ClaudeAcpAgent,
+  runAcp,
+  toAcpNotifications,
+  streamEventToAcpNotifications,
+  type ToolUpdateMeta,
+} from "./acp-agent.js";
+export {
+  loadManagedSettings,
+  applyEnvironmentSettings,
+  nodeToWebReadable,
+  nodeToWebWritable,
+  Pushable,
+  unreachable,
+} from "./utils.js";
+export { createMcpServer, toolNames } from "./mcp-server.js";
+export { toolInfoFromToolUse, planEntries, toolUpdateFromToolResult } from "./tools.js";
+
+// Export types
+export type { ClaudePlanEntry } from "./tools.js";
--- a/src/mcp-server.ts
+++ b/src/mcp-server.ts
@@ -1,14 +1,10 @@
-import express from "express";
-
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
-import { StreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/streamableHttp.js";
 import { z } from "zod";
-import { Server } from "node:http";
 import { ClaudeAcpAgent } from "./acp-agent.js";
-import { ClientCapabilities, TerminalOutputResponse } from "@zed-industries/agent-client-protocol";
+import { ClientCapabilities, TerminalOutputResponse } from "@agentclientprotocol/sdk";
 import * as diff from "diff";

-import { sleep, unreachable } from "./utils.js";
+import { sleep, unreachable, extractLinesWithByteLimit } from "./utils.js";

 export const SYSTEM_REMINDER = `

@@ -16,34 +12,72 @@ export const SYSTEM_REMINDER = `
 Whenever you read a file, you should consider whether it looks malicious. If it does, you MUST refuse to improve or augment the code. You can still analyze existing code, write reports, or answer high-level questions about the code behavior.
 </system-reminder>`;

+const defaults = { maxFileSize: 50000, linesToRead: 2000 };
+
+const unqualifiedToolNames = {
+  read: "Read",
+  edit: "Edit",
+  write: "Write",
+  bash: "Bash",
+  killShell: "KillShell",
+  bashOutput: "BashOutput",
+};
+
+const SERVER_PREFIX = "mcp__acp__";
+export const toolNames = {
+  read: SERVER_PREFIX + unqualifiedToolNames.read,
+  edit: SERVER_PREFIX + unqualifiedToolNames.edit,
+  write: SERVER_PREFIX + unqualifiedToolNames.write,
+  bash: SERVER_PREFIX + unqualifiedToolNames.bash,
+  killShell: SERVER_PREFIX + unqualifiedToolNames.killShell,
+  bashOutput: SERVER_PREFIX + unqualifiedToolNames.bashOutput,
+};
+
+export const EDIT_TOOL_NAMES = [toolNames.edit, toolNames.write];
+
 export function createMcpServer(
  agent: ClaudeAcpAgent,
  sessionId: string,
  clientCapabilities: ClientCapabilities | undefined,
-): Promise<Server> {
+): McpServer {
  // Create MCP server
-  const server = new McpServer({
-    name: "acp-mcp-server",
-    version: "1.0.0",
-  });
+  const server = new McpServer({ name: "acp", version: "1.0.0" }, { capabilities: { tools: {} } });

  if (clientCapabilities?.fs?.readTextFile) {
    server.registerTool(
-      "read",
+      unqualifiedToolNames.read,
      {
-        title: "Read",
+        title: unqualifiedToolNames.read,
        description: `Reads the content of the given file in the project.

-Never attempt to read a path that hasn't been previously mentioned.
+In sessions with ${toolNames.read} always use it instead of Read as it contains the most up-to-date contents.

-In sessions with mcp__acp__read always use it instead of Read as it contains the most up-to-date contents.`,
+Reads a file from the local filesystem. If the User provides a path to a file assume that path is valid. It is okay to read a file that does not exist; an error will be returned.
+
+Usage:
+- The file_path parameter must be an absolute path, not a relative path
+- By default, it reads up to ${defaults.linesToRead} lines starting from the beginning of the file
+- You can optionally specify a line offset and limit (especially handy for long files), but it's recommended to read the whole file by not providing these parameters
+- Any files larger than ${defaults.maxFileSize} bytes will be truncated
+- This tool allows Claude Code to read images (eg PNG, JPG, etc). When reading an image file the contents are presented visually as Claude Code is a multimodal LLM.
+- This tool can only read files, not directories. To read a directory, use an ls command via the ${toolNames.bash} tool.
+- You have the capability to call multiple tools in a single response. It is always better to speculatively read multiple files as a batch that are potentially useful.`,
        inputSchema: {
-          abs_path: z.string().describe("The absolute path to the file to read."),
+          file_path: z.string().describe("The absolute path to the file to read"),
          offset: z
            .number()
            .optional()
-            .describe("Which line to start reading from. Omit to start from the beginning."),
-          limit: z.number().optional().describe("How many lines to read. Omit for the whole file."),
+            .default(1)
+            .describe(
+              "The line number to start reading from. Only provide if the file is too large to read at once",
+            ),
+          limit: z
+            .number()
+            .optional()
+            .default(defaults.linesToRead)
+            .describe(
+              `The number of lines to read. Only provide if the file is too large to read at once.`,
+            ),
        },
        annotations: {
          title: "Read file",
@@ -66,18 +100,40 @@ In sessions with mcp__acp__read always use it instead of Read as it contains the
              ],
            };
          }
+
          const content = await agent.readTextFile({
            sessionId,
-            path: input.abs_path,
-            limit: input.limit,
+            path: input.file_path,
            line: input.offset,
+            limit: input.limit,
          });

+          // Extract lines with byte limit enforcement
+          const result = extractLinesWithByteLimit(content.content, defaults.maxFileSize);
+
+          // Construct informative message about what was read
+          let readInfo = "";
+          if (input.offset > 1 || result.wasLimited) {
+            readInfo = "\n\n<file-read-info>";
+
+            if (result.wasLimited) {
+              readInfo += `Read ${result.linesRead} lines (hit 50KB limit). `;
+            } else {
+              readInfo += `Read lines ${input.offset}-${result.linesRead}. `;
+            }
+
+            if (result.wasLimited) {
+              readInfo += `Continue with offset=${result.linesRead}.`;
+            }
+
+            readInfo += "</file-read-info>";
+          }
+
          return {
            content: [
              {
                type: "text",
-                text: content.content + SYSTEM_REMINDER,
+                text: result.content + readInfo + SYSTEM_REMINDER,
              },
            ],
          };
@@ -97,16 +153,25 @@ In sessions with mcp__acp__read always use it instead of Read as it contains the

  if (clientCapabilities?.fs?.writeTextFile) {
    server.registerTool(
-      "write",
+      unqualifiedToolNames.write,
      {
-        title: "Write",
-        description: `Writes content to the specified file in the project.
+        title: unqualifiedToolNames.write,
+        description: `Writes a file to the local filesystem..

-In sessions with mcp__acp__write always use it instead of Write as it will
-allow the user to conveniently review changes.`,
+In sessions with ${toolNames.write} always use it instead of Write as it will
+allow the user to conveniently review changes.
+
+Usage:
+- This tool will overwrite the existing file if there is one at the provided path.
+- If this is an existing file, you MUST use the ${toolNames.read} tool first to read the file's contents. This tool will fail if you did not read the file first.
+- ALWAYS prefer editing existing files in the codebase. NEVER write new files unless explicitly required.
+- NEVER proactively create documentation files (*.md) or README files. Only create documentation files if explicitly requested by the User.
+- Only use emojis if the user explicitly requests it. Avoid writing emojis to files unless asked.`,
        inputSchema: {
-          abs_path: z.string().describe("The absolute path to the file to write"),
-          content: z.string().describe("The full content to write"),
+          file_path: z
+            .string()
+            .describe("The absolute path to the file to write (must be absolute, not relative)"),
+          content: z.string().describe("The content to write to the file"),
        },
        annotations: {
          title: "Write file",
@@ -131,7 +196,7 @@ allow the user to conveniently review changes.`,
          }
          await agent.writeTextFile({
            sessionId,
-            path: input.abs_path,
+            path: input.file_path,
            content: input.content,
          });

@@ -152,28 +217,32 @@ allow the user to conveniently review changes.`,
    );

    server.registerTool(
-      "edit",
+      unqualifiedToolNames.edit,
      {
-        title: "Edit",
-        description: `Edit a file.
+        title: unqualifiedToolNames.edit,
+        description: `Performs exact string replacements in files.

-In sessions with mcp__acp__edit always use it instead of Edit as it will
+In sessions with ${toolNames.edit} always use it instead of Edit as it will
 allow the user to conveniently review changes.

-File editing instructions:
- The \`old_string\` param must match existing file content, including indentation.
- The \`old_string\` param must come from the actual file, not an outline.
- The \`old_string\` section must not be empty.
- Be minimal with replacements:
-  - For unique lines, include only those lines.
-  - For non-unique lines, include enough context to identify them.
- Do not escape quotes, newlines, or other characters.
- Only edit the specified file.
- If the \`old_string\` value isn't found in the file, the edit won't be applied. The tool will fail and must be retried.`,
+Usage:
+- You must use your \`${toolNames.read}\` tool at least once in the conversation before editing. This tool will error if you attempt an edit without reading the file.
+- When editing text from Read tool output, ensure you preserve the exact indentation (tabs/spaces) as it appears.
+- ALWAYS prefer editing existing files in the codebase. NEVER write new files unless explicitly required.
+- Only use emojis if the user explicitly requests it. Avoid adding emojis to files unless asked.
+- The edit will FAIL if \`old_string\` is not unique in the file. Either provide a larger string with more surrounding context to make it unique or use \`replace_all\` to change every instance of \`old_string\`.
+- Use \`replace_all\` for replacing and renaming strings across the file. This parameter is useful if you want to rename a variable for instance.`,
        inputSchema: {
-          abs_path: z.string().describe("The absolute path to the file to read."),
-          old_string: z.string().describe("The old text to replace (must be unique in the file)"),
-          new_string: z.string().describe("The new text."),
+          file_path: z.string().describe("The absolute path to the file to modify"),
+          old_string: z.string().describe("The text to replace"),
+          new_string: z
+            .string()
+            .describe("The text to replace it with (must be different from old_string)"),
+          replace_all: z
+            .boolean()
+            .default(false)
+            .optional()
+            .describe("Replace all occurences of old_string (default false)"),
        },
        annotations: {
          title: "Edit file",
@@ -198,105 +267,17 @@ File editing instructions:

        const { content } = await agent.readTextFile({
          sessionId,
-          path: input.abs_path,
+          path: input.file_path,
        });

        const { newContent } = replaceAndCalculateLocation(content, [
          {
            oldText: input.old_string,
            newText: input.new_string,
-            replaceAll: false,
+            replaceAll: input.replace_all,
          },
        ]);

-        const patch = diff.createPatch(input.abs_path, content, newContent);
-
-        await agent.writeTextFile({
-          sessionId,
-          path: input.abs_path,
-          content: newContent,
-        });
-
-        return {
-          content: [
-            {
-              type: "text",
-              text: patch,
-            },
-          ],
-        };
-      },
-    );
-
-    server.registerTool(
-      "multi-edit",
-      {
-        title: "Multi Edit",
-        description: `Edit a file with multiple sequential edits.
-
-In sessions with mcp__acp__multi-edit always use it instead of MultiEdit as it will
-allow the user to conveniently review changes.
-
-File editing instructions:
- The \`old_string\` param must match existing file content, including indentation.
- The \`old_string\` param must come from the actual file, not an outline.
- The \`old_string\` section must not be empty.
- Be minimal with replacements:
-  - For unique lines, include only those lines.
-  - For non-unique lines, include enough context to identify them, unless you're using \`replace_all\`
- Do not escape quotes, newlines, or other characters.
- If any of the provided \`old_string\` values aren't found in the file, no edits will be applied. The tool will fail and must be retried.`,
-        inputSchema: {
-          file_path: z.string().describe("The absolute path to the file to modify"),
-          edits: z
-            .array(
-              z.object({
-                old_string: z.string().describe("The text to replace"),
-                new_string: z.string().describe("The text to replace it with"),
-                replace_all: z
-                  .boolean()
-                  .optional()
-                  .describe("Replace all occurrences of old_string (default false)"),
-              }),
-            )
-            .min(1)
-            .describe("Array of edit operations to perform sequentially on the file"),
-        },
-        annotations: {
-          title: "Multi Edit file",
-          readOnlyHint: false,
-          destructiveHint: false,
-          openWorldHint: false,
-          idempotentHint: false,
-        },
-      },
-      async (input) => {
-        const session = agent.sessions[sessionId];
-        if (!session) {
-          return {
-            content: [
-              {
-                type: "text",
-                text: "The user has left the building",
-              },
-            ],
-          };
-        }
-
-        const { content } = await agent.readTextFile({
-          sessionId,
-          path: input.file_path,
-        });
-
-        const { newContent } = replaceAndCalculateLocation(
-          content,
-          input.edits.map((edit) => ({
-            oldText: edit.old_string,
-            newText: edit.new_string,
-            replaceAll: edit.replace_all ?? false,
-          })),
-        );
-
        const patch = diff.createPatch(input.file_path, content, newContent);

        await agent.writeTextFile({
@@ -319,21 +300,36 @@ File editing instructions:

  if (agent.clientCapabilities?.terminal) {
    server.registerTool(
-      "Bash",
+      unqualifiedToolNames.bash,
      {
-        title: "Bash",
-        description: "Executes a bash command",
+        title: unqualifiedToolNames.bash,
+        description: `Executes a bash command
+
+In sessions with ${toolNames.bash} always use it instead of Bash`,
        inputSchema: {
-          command: z.string().describe("The bash command to execute as a one-liner"),
-          timeout_ms: z
+          command: z.string().describe("The command to execute"),
+          timeout: z
            .number()
            .default(2 * 60 * 1000)
-            .describe("Optional timeout in milliseconds"),
+            .describe(`Optional timeout in milliseconds (max ${2 * 60 * 1000})`),
+          description: z.string().optional()
+            .describe(`Clear, concise description of what this command does in 5-10 words, in active voice. Examples:
+Input: ls
+Output: List files in current directory
+
+Input: git status
+Output: Show working tree status
+
+Input: npm install
+Output: Install package dependencies
+
+Input: mkdir foo
+Output: Create directory 'foo'`),
          run_in_background: z
            .boolean()
            .default(false)
            .describe(
-              "When set to true, the command is started in the background. The tool returns an `id` that can be used with the `mcp__acp__BashOutput` tool to retrieve the current output, or the `mcp__acp__KillBash` tool to stop it early.",
+              `Set to true to run this command in the background. The tool returns an \`id\` that can be used with the \`${toolNames.bashOutput}\` tool to retrieve the current output, or the \`${toolNames.killShell}\` tool to stop it early.`,
            ),
        },
      },
@@ -372,6 +368,7 @@ File editing instructions:
            sessionUpdate: "tool_call_update",
            toolCallId,
            status: "in_progress",
+            title: input.description,
            content: [{ type: "terminal", terminalId: handle.id }],
          },
        });
@@ -389,7 +386,7 @@ File editing instructions:
        const statusPromise = Promise.race([
          handle.waitForExit().then((exitStatus) => ({ status: "exited" as const, exitStatus })),
          abortPromise.then(() => ({ status: "aborted" as const, exitStatus: null })),
-          sleep(input.timeout_ms).then(async () => {
+          sleep(input.timeout).then(async () => {
            if (agent.backgroundTerminals[handle.id]?.status === "started") {
              await handle.kill();
            }
@@ -454,22 +451,27 @@ File editing instructions:
    );

    server.registerTool(
-      "BashOutput",
+      unqualifiedToolNames.bashOutput,
      {
-        title: "BashOutput",
-        description:
-          "Returns the current output and exit status of a background bash command by its id. Includes only new output since last invocation.",
+        title: unqualifiedToolNames.bashOutput,
+        description: `- Retrieves output from a running or completed background bash shell
+- Takes a shell_id parameter identifying the shell
+- Always returns only new output since the last check
+- Returns stdout and stderr output along with shell status
+- Use this tool when you need to monitor or check the output of a long-running shell
+
+In sessions with ${toolNames.bashOutput} always use it instead of BashOutput.`,
        inputSchema: {
-          id: z
+          shell_id: z
            .string()
-            .describe("The id of the background bash command as returned by `mcp__acp__Bash`"),
+            .describe(`The id of the background bash command as returned by \`${toolNames.bash}\``),
        },
      },
      async (input) => {
-        const bgTerm = agent.backgroundTerminals[input.id];
+        const bgTerm = agent.backgroundTerminals[input.shell_id];

        if (!bgTerm) {
-          throw new Error(`Unknown shell ${input.id}`);
+          throw new Error(`Unknown shell ${input.shell_id}`);
        }

        if (bgTerm.status === "started") {
@@ -505,21 +507,26 @@ File editing instructions:
    );

    server.registerTool(
-      "KillBash",
+      unqualifiedToolNames.killShell,
      {
-        title: "KillBash",
-        description: "Stops a background command by its id",
+        title: unqualifiedToolNames.killShell,
+        description: `- Kills a running background bash shell by its ID
+- Takes a shell_id parameter identifying the shell to kill
+- Returns a success or failure status
+- Use this tool when you need to terminate a long-running shell
+
+In sessions with ${toolNames.killShell} always use it instead of KillShell.`,
        inputSchema: {
-          id: z
+          shell_id: z
            .string()
-            .describe("The id of the background bash command as returned by `mcp__acp__Bash`"),
+            .describe(`The id of the background bash command as returned by \`${toolNames.bash}\``),
        },
      },
      async (input) => {
-        const bgTerm = agent.backgroundTerminals[input.id];
+        const bgTerm = agent.backgroundTerminals[input.shell_id];

        if (!bgTerm) {
-          throw new Error(`Unknown shell ${input.id}`);
+          throw new Error(`Unknown shell ${input.shell_id}`);
        }

        switch (bgTerm.status) {
@@ -563,133 +570,7 @@ File editing instructions:
    );
  }

-  const alwaysAllowedTools: { [key: string]: boolean } = {};
-  server.registerTool(
-    "permission",
-    {
-      title: "Permission Tool",
-      description: "Used to request tool permissions",
-      inputSchema: {
-        tool_name: z.string(),
-        input: z.any(),
-        tool_use_id: z.string().optional(),
-      },
-    },
-    async (input) => {
-      const session = agent.sessions[sessionId];
-      if (!session) {
-        return {
-          content: [
-            {
-              type: "text",
-              text: JSON.stringify({
-                behavior: "deny",
-                message: "Session not found",
-              }),
-            },
-          ],
-        };
-      }
-      if (alwaysAllowedTools[input.tool_name]) {
-        return {
-          content: [
-            {
-              type: "text",
-              text: JSON.stringify({
-                behavior: "allow",
-                updatedInput: input.input,
-              }),
-            },
-          ],
-        };
-      }
-      const response = await agent.client.requestPermission({
-        options: [
-          {
-            kind: "allow_always",
-            name: "Always Allow",
-            optionId: "allow_always",
-          },
-          { kind: "allow_once", name: "Allow", optionId: "allow" },
-          { kind: "reject_once", name: "Reject", optionId: "reject" },
-        ],
-        sessionId,
-        toolCall: {
-          toolCallId: input.tool_use_id!,
-          rawInput: input.input,
-        },
-      });
-      if (
-        response.outcome?.outcome === "selected" &&
-        (response.outcome.optionId === "allow" || response.outcome.optionId === "allow_always")
-      ) {
-        if (response.outcome.optionId === "allow_always") {
-          alwaysAllowedTools[input.tool_name] = true;
-        }
-        return {
-          content: [
-            {
-              type: "text",
-              text: JSON.stringify({
-                behavior: "allow",
-                updatedInput: input.input,
-              }),
-            },
-          ],
-        };
-      } else {
-        return {
-          content: [
-            {
-              type: "text",
-              text: JSON.stringify({
-                behavior: "deny",
-                message: "User refused permission to run tool",
-              }),
-            },
-          ],
-        };
-      }
-    },
-  );
-
-  const app = express();
-  app.use(express.json());
-
-  app.post("/mcp", async (req, res) => {
-    try {
-      const transport: StreamableHTTPServerTransport = new StreamableHTTPServerTransport({
-        sessionIdGenerator: undefined,
-      });
-      res.on("close", () => {
-        transport.close();
-        server.close();
-      });
-      await server.connect(transport);
-      await transport.handleRequest(req, res, req.body);
-    } catch (error) {
-      if (!res.headersSent) {
-        res.status(500).json({
-          jsonrpc: "2.0",
-          error: {
-            code: -32603,
-            message: `Internal server error: ${error}`,
-          },
-          id: null,
-        });
-      }
-    }
-  });
-
-  return new Promise((resolve, reject) => {
-    const listener = app.listen(0, "127.0.0.1", (error) => {
-      if (error) {
-        reject(error);
-        return;
-      }
-      resolve(listener);
-    });
-  });
+  return server;
 }

 function stripCommonPrefix(a: string, b: string): string {
--- a/src/tests/acp-agent.test.ts
+++ b/src/tests/acp-agent.test.ts
@@ -2,9 +2,11 @@ import { describe, it, expect, beforeAll, afterAll } from "vitest";
 import { spawn, spawnSync } from "child_process";
 import {
  Agent,
+  AgentSideConnection,
  AvailableCommand,
  Client,
  ClientSideConnection,
+  ndJsonStream,
  NewSessionResponse,
  ReadTextFileRequest,
  ReadTextFileResponse,
@@ -13,11 +15,11 @@ import {
  SessionNotification,
  WriteTextFileRequest,
  WriteTextFileResponse,
-} from "@zed-industries/agent-client-protocol";
+} from "@agentclientprotocol/sdk";
 import { nodeToWebWritable, nodeToWebReadable } from "../utils.js";
 import { markdownEscape, toolInfoFromToolUse, toolUpdateFromToolResult } from "../tools.js";
-import { toAcpNotifications } from "../acp-agent.js";
-import { SDKAssistantMessage } from "@anthropic-ai/claude-code";
+import { toAcpNotifications, promptToClaude } from "../acp-agent.js";
+import { query, SDKAssistantMessage } from "@anthropic-ai/claude-agent-sdk";

 describe.skipIf(!process.env.RUN_INTEGRATION_TESTS)("ACP subprocess integration", () => {
  let child: ReturnType<typeof spawn>;
@@ -91,7 +93,7 @@ describe.skipIf(!process.env.RUN_INTEGRATION_TESTS)("ACP subprocess integration"

    async writeTextFile(params: WriteTextFileRequest): Promise<WriteTextFileResponse> {
      this.files.set(params.path, params.content);
-      return null;
+      return {};
    }

    async readTextFile(params: ReadTextFileRequest): Promise<ReadTextFileResponse> {
@@ -108,14 +110,13 @@ describe.skipIf(!process.env.RUN_INTEGRATION_TESTS)("ACP subprocess integration"
    newSessionResponse: NewSessionResponse;
  }> {
    let client;
-    const connection = new ClientSideConnection(
-      (agent) => {
-        client = new TestClient(agent);
-        return client;
-      },
-      nodeToWebWritable(child.stdin!),
-      nodeToWebReadable(child.stdout!),
-    );
+    const input = nodeToWebWritable(child.stdin!);
+    const output = nodeToWebReadable(child.stdout!);
+    const stream = ndJsonStream(input, output);
+    const connection = new ClientSideConnection((agent) => {
+      client = new TestClient(agent);
+      return client;
+    }, stream);

    await connection.initialize({
      protocolVersion: 1,
@@ -191,6 +192,50 @@ describe.skipIf(!process.env.RUN_INTEGRATION_TESTS)("ACP subprocess integration"

    expect(client.takeReceivedText()).toContain("Hello GPT-5");
  }, 30000);
+
+  it("/compact works", async () => {
+    const { client, connection, newSessionResponse } = await setupTestSession(__dirname);
+
+    const commands = await client.availableCommandsPromise;
+
+    expect(commands).toContainEqual({
+      description:
+        "Clear conversation history but keep a summary in context. Optional: /compact [instructions for summarization]",
+      input: {
+        hint: "<optional custom summarization instructions>",
+      },
+      name: "compact",
+    });
+
+    // Error case (no previous message)
+    await connection.prompt({
+      prompt: [{ type: "text", text: "/compact" }],
+      sessionId: newSessionResponse.sessionId,
+    });
+
+    expect(client.takeReceivedText()).toBe("");
+
+    // Send something
+    await connection.prompt({
+      prompt: [{ type: "text", text: "Hi" }],
+      sessionId: newSessionResponse.sessionId,
+    });
+    // Clear response
+    client.takeReceivedText();
+
+    // Test with instruction
+    await connection.prompt({
+      prompt: [
+        {
+          type: "text",
+          text: "/compact greeting",
+        },
+      ],
+      sessionId: newSessionResponse.sessionId,
+    });
+
+    expect(client.takeReceivedText()).toContain("");
+  }, 30000);
 });

 describe("tool conversions", () => {
@@ -327,13 +372,13 @@ describe("tool conversions", () => {
    });
  });

-  it("should handle mcp__acp__write tool calls", () => {
+  it("should handle mcp__acp__Write tool calls", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01GHI789JKL456",
-      name: "mcp__acp__write",
+      name: "mcp__acp__Write",
      input: {
-        abs_path: "/Users/test/project/config.json",
+        file_path: "/Users/test/project/config.json",
        content: '{"version": "1.0.0"}',
      },
    };
@@ -371,13 +416,13 @@ describe("tool conversions", () => {
    });
  });

-  it("should handle mcp__acp__read tool calls", () => {
+  it("should handle mcp__acp__Read tool calls", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01YZA789BCD123",
-      name: "mcp__acp__read",
+      name: "mcp__acp__Read",
      input: {
-        abs_path: "/Users/test/project/data.json",
+        file_path: "/Users/test/project/data.json",
      },
    };

@@ -389,13 +434,13 @@ describe("tool conversions", () => {
    });
  });

-  it("should handle mcp__acp__read with limit", () => {
+  it("should handle mcp__acp__Read with limit", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01EFG456HIJ789",
-      name: "mcp__acp__read",
+      name: "mcp__acp__Read",
      input: {
-        abs_path: "/Users/test/project/large.txt",
+        file_path: "/Users/test/project/large.txt",
        limit: 100,
      },
    };
@@ -408,13 +453,13 @@ describe("tool conversions", () => {
    });
  });

-  it("should handle mcp__acp__read with offset and limit", () => {
+  it("should handle mcp__acp__Read with offset and limit", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01KLM789NOP456",
-      name: "mcp__acp__read",
+      name: "mcp__acp__Read",
      input: {
-        abs_path: "/Users/test/project/large.txt",
+        file_path: "/Users/test/project/large.txt",
        offset: 50,
        limit: 100,
      },
@@ -428,13 +473,13 @@ describe("tool conversions", () => {
    });
  });

-  it("should handle mcp__acp__read with only offset", () => {
+  it("should handle mcp__acp__Read with only offset", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01QRS123TUV789",
-      name: "mcp__acp__read",
+      name: "mcp__acp__Read",
      input: {
-        abs_path: "/Users/test/project/large.txt",
+        file_path: "/Users/test/project/large.txt",
        offset: 200,
      },
    };
@@ -495,7 +540,7 @@ describe("tool conversions", () => {
    const tool_use = {
      type: "tool_use",
      id: "toolu_01PhLms5fuvmdjy2bb6dfUKT",
-      name: "KillBash",
+      name: "KillShell",
      input: {
        shell_id: "bash_1",
      },
@@ -532,6 +577,7 @@ describe("tool conversions", () => {
        id: "msg_017eNosJgww7F5qD4a8BcAcx",
        type: "message",
        role: "assistant",
+        container: null,
        model: "claude-sonnet-4-20250514",
        content: [
          {
@@ -596,13 +642,24 @@ describe("tool conversions", () => {
          },
          output_tokens: 1,
          service_tier: "standard",
+          server_tool_use: null,
        },
+        context_management: null,
      },
      parent_tool_use_id: null,
      session_id: "d056596f-e328-41e9-badd-b07122ae5227",
      uuid: "b7c3330c-de8f-4bba-ac53-68c7f76ffeb5",
    };
-    expect(toAcpNotifications(received, "test", {}, {})).toStrictEqual([
+    expect(
+      toAcpNotifications(
+        received.message.content,
+        received.message.role,
+        "test",
+        {},
+        {},
+        {} as AgentSideConnection,
+      ),
+    ).toStrictEqual([
      {
        sessionId: "test",
        update: {
@@ -654,78 +711,13 @@ describe("tool conversions", () => {
    ]);
  });

-  it("should show full diff for multi edit tool", () => {
-    const toolUse = {
-      type: "tool_use",
-      id: "toolu_01DEF456",
-      name: "mcp__acp__multi-edit",
-      input: {
-        file_path: "/Users/test/project/config.json",
-        edits: [
-          {
-            old_string: 'version": 1',
-            new_string: 'version": 2',
-            replace_all: false,
-          },
-          {
-            old_string: 'enabled": false',
-            new_string: 'enabled": true',
-            replace_all: true,
-          },
-        ],
-      },
-    };
-    const fileCache: { [key: string]: string } = {
-      "/Users/test/project/config.json": JSON.stringify(
-        {
-          version: 1,
-          filler: "filler",
-          enabled: false,
-        },
-        null,
-        4,
-      ),
-    };
-
-    expect(toolInfoFromToolUse(toolUse, fileCache)).toEqual({
-      content: [
-        {
-          newText: `{
-    "version": 2,
-    "filler": "filler",
-    "enabled": true
-}`,
-          oldText: `{
-    "version": 1,
-    "filler": "filler",
-    "enabled": false
-}`,
-          path: "/Users/test/project/config.json",
-          type: "diff",
-        },
-      ],
-      kind: "edit",
-      locations: [
-        {
-          path: "/Users/test/project/config.json",
-          line: 1,
-        },
-        {
-          path: "/Users/test/project/config.json",
-          line: 3,
-        },
-      ],
-      title: "Edit /Users/test/project/config.json",
-    });
-  });
-
  it("should return empty update for successful edit result", () => {
    const toolUse = {
      type: "tool_use",
      id: "toolu_01MNO345",
-      name: "mcp__acp__edit",
+      name: "mcp__acp__Edit",
      input: {
-        abs_path: "/Users/test/project/test.txt",
+        file_path: "/Users/test/project/test.txt",
        old_string: "old",
        new_string: "new",
      },
@@ -734,7 +726,7 @@ describe("tool conversions", () => {
    const toolResult = {
      content: [
        {
-          type: "text",
+          type: "text" as const,
          text: "not valid json",
        },
      ],
@@ -753,9 +745,9 @@ describe("tool conversions", () => {
    const toolUse = {
      type: "tool_use",
      id: "toolu_01MNO345",
-      name: "mcp__acp__edit",
+      name: "mcp__acp__Edit",
      input: {
-        abs_path: "/Users/test/project/test.txt",
+        file_path: "/Users/test/project/test.txt",
        old_string: "old",
        new_string: "new",
      },
@@ -764,7 +756,7 @@ describe("tool conversions", () => {
    const toolResult = {
      content: [
        {
-          type: "text",
+          type: "text" as const,
          text: "Failed to find `old_string`",
        },
      ],
@@ -778,7 +770,10 @@ describe("tool conversions", () => {
    // Should return empty object when parsing fails
    expect(update).toEqual({
      content: [
-        { content: { type: "text", text: "Failed to find `old_string`" }, type: "content" },
+        {
+          content: { type: "text", text: "```\nFailed to find `old_string`\n```" },
+          type: "content",
+        },
      ],
    });
  });
@@ -795,3 +790,54 @@ describe("escape markdown", () => {
    expect(escaped).toEqual("````\nfor example:\n```markdown\nHello *world*!\n```\n````");
  });
 });
+
+describe("prompt conversion", () => {
+  it("should not change built-in slash commands", () => {
+    const message = promptToClaude({
+      sessionId: "test",
+      prompt: [
+        {
+          type: "text",
+          text: "/compact args",
+        },
+      ],
+    });
+    expect(message.message.content).toEqual([
+      {
+        text: "/compact args",
+        type: "text",
+      },
+    ]);
+  });
+
+  it("should remove MCP prefix from MCP slash commands", () => {
+    const message = promptToClaude({
+      sessionId: "test",
+      prompt: [
+        {
+          type: "text",
+          text: "/mcp:server:name args",
+        },
+      ],
+    });
+    expect(message.message.content).toEqual([
+      {
+        text: "/server:name (MCP) args",
+        type: "text",
+      },
+    ]);
+  });
+});
+
+describe("SDK behavior", () => {
+  it.skipIf(!process.env.RUN_INTEGRATION_TESTS)(
+    "query has a 'default' model",
+    async () => {
+      const q = query({ prompt: "hi" });
+      const models = await q.supportedModels();
+      const defaultModel = models.find((m) => m.value === "default");
+      expect(defaultModel).toBeDefined();
+    },
+    10000,
+  );
+});
--- a/src/tests/extract-lines.test.ts
+++ b/src/tests/extract-lines.test.ts
@@ -0,0 +1,103 @@
+import { describe, it, expect } from "vitest";
+import { extractLinesWithByteLimit } from "../utils.js";
+
+describe("extractLinesWithByteLimit", () => {
+  const simpleContent = "Line 1\nLine 2\nLine 3\nLine 4\nLine 5";
+
+  it("should extract all lines when under byte limit", () => {
+    const result = extractLinesWithByteLimit(simpleContent, 1000);
+
+    expect(result.content).toBe(simpleContent);
+    expect(result.wasLimited).toBe(false);
+    expect(result.linesRead).toBe(5);
+  });
+
+  it("should limit output when exceeding byte limit", () => {
+    // Create content that will exceed byte limit
+    const longLine = "x".repeat(100);
+    const manyLines = Array(10).fill(longLine).join("\n");
+
+    const result = extractLinesWithByteLimit(manyLines, 250);
+
+    expect(result.wasLimited).toBe(true);
+    expect(result.linesRead).toBe(2);
+  });
+
+  it("should handle empty content", () => {
+    const result = extractLinesWithByteLimit("", 1000);
+
+    expect(result.content).toBe("");
+    expect(result.wasLimited).toBe(false);
+    expect(result.linesRead).toBe(1); // We read the one empty line
+  });
+
+  it("should handle single line file", () => {
+    const singleLine = "This is a single line without newline";
+    const result = extractLinesWithByteLimit(singleLine, 1000);
+
+    expect(result.content).toBe(singleLine);
+    expect(result.wasLimited).toBe(false);
+    expect(result.linesRead).toBe(1);
+  });
+
+  it("should correctly count bytes with multi-byte characters", () => {
+    const unicodeContent = "Hello 世界\n你好 World\nEmoji: 🌍\nNormal line";
+    const result = extractLinesWithByteLimit(unicodeContent, 1000);
+
+    expect(result.content).toBe(unicodeContent);
+    expect(result.linesRead).toBe(4);
+  });
+
+  it("should stop at byte limit even with one more line available", () => {
+    // Create lines where adding one more would exceed limit
+    const line1 = "a".repeat(40);
+    const line2 = "b".repeat(40);
+    const line3 = "c".repeat(40);
+    const content = `${line1}\n${line2}\n${line3}`;
+
+    const result = extractLinesWithByteLimit(content, 85);
+
+    expect(result.content).toBe(`${line1}\n${line2}\n`);
+    expect(result.wasLimited).toBe(true);
+    expect(result.linesRead).toBe(2);
+  });
+
+  it("should read exactly to limit when possible", () => {
+    const exactContent = "12345\n67890\n12345"; // 17 bytes total
+    const result = extractLinesWithByteLimit(exactContent, 17);
+
+    expect(result.content).toBe(exactContent);
+    expect(result.wasLimited).toBe(false);
+    expect(result.linesRead).toBe(3);
+  });
+
+  it("should handle Windows-style line endings", () => {
+    const windowsContent = "Line 1\r\nLine 2\r\nLine 3";
+    const result = extractLinesWithByteLimit(windowsContent, 1000);
+
+    // Note: split("\n") will keep the \r characters
+    expect(result.content).toBe("Line 1\r\nLine 2\r\nLine 3");
+    expect(result.linesRead).toBe(3);
+  });
+
+  it("should handle very large files efficiently", () => {
+    // Create a 100KB file
+    const largeLine = "x".repeat(1000);
+    const largeContent = Array(110).fill(largeLine).join("\n");
+
+    const result = extractLinesWithByteLimit(largeContent, 50000);
+
+    expect(result.wasLimited).toBe(true);
+  });
+
+  it("should allow at least one line even if it exceeds byte limit", () => {
+    const veryLongLine = "x".repeat(100000); // 100KB line
+    const result = extractLinesWithByteLimit(veryLongLine, 50000);
+
+    // Should return the line even though it exceeds the byte limit
+    // because we always allow at least one line if no lines have been added yet
+    expect(result.content).toBe(veryLongLine);
+    expect(result.linesRead).toBe(1);
+    expect(result.wasLimited).toBe(false);
+  });
+});
--- a/src/tools.ts
+++ b/src/tools.ts
@@ -1,10 +1,15 @@
+import { PlanEntry, ToolCallContent, ToolCallLocation, ToolKind } from "@agentclientprotocol/sdk";
+import { replaceAndCalculateLocation, SYSTEM_REMINDER, toolNames } from "./mcp-server.js";
+import { ToolResultBlockParam, WebSearchToolResultBlockParam } from "@anthropic-ai/sdk/resources";
 import {
-  PlanEntry,
-  ToolCallContent,
-  ToolCallLocation,
-  ToolKind,
-} from "@zed-industries/agent-client-protocol";
-import { replaceAndCalculateLocation, SYSTEM_REMINDER } from "./mcp-server.js";
+  BetaBashCodeExecutionToolResultBlockParam,
+  BetaCodeExecutionToolResultBlockParam,
+  BetaRequestMCPToolResultBlockParam,
+  BetaTextEditorCodeExecutionToolResultBlockParam,
+  BetaWebFetchToolResultBlockParam,
+  BetaWebSearchToolResultBlockParam,
+} from "@anthropic-ai/sdk/resources/beta.mjs";
+import { HookCallback } from "@anthropic-ai/claude-agent-sdk";

 interface ToolInfo {
  title: string;
@@ -67,7 +72,7 @@ export function toolInfoFromToolUse(
      };

    case "Bash":
-    case "mcp__acp__Bash":
+    case toolNames.bash:
      return {
        title: input?.command ? "`" + input.command.replaceAll("`", "\\`") + "`" : "Terminal",
        kind: "execute",
@@ -83,22 +88,22 @@ export function toolInfoFromToolUse(
      };

    case "BashOutput":
-    case "mcp__acp__BashOutput":
+    case toolNames.bashOutput:
      return {
        title: "Tail Logs",
        kind: "execute",
        content: [],
      };

-    case "KillBash":
-    case "mcp__acp__KillBash":
+    case "KillShell":
+    case toolNames.killShell:
      return {
        title: "Kill Process",
        kind: "execute",
        content: [],
      };

-    case "mcp__acp__read": {
+    case toolNames.read: {
      let limit = "";
      if (input.limit) {
        limit =
@@ -107,14 +112,16 @@ export function toolInfoFromToolUse(
        limit = " (from line " + (input.offset + 1) + ")";
      }
      return {
-        title: "Read " + (input.abs_path ?? "File") + limit,
+        title: "Read " + (input.file_path ?? "File") + limit,
        kind: "read",
-        locations: [
-          {
-            path: input.abs_path,
-            line: input.offset ?? 0,
-          },
-        ],
+        locations: input.file_path
+          ? [
+              {
+                path: input.file_path,
+                line: input.offset ?? 0,
+              },
+            ]
+          : [],
        content: [],
      };
    }
@@ -135,9 +142,9 @@ export function toolInfoFromToolUse(
        locations: [],
      };

-    case "mcp__acp__edit":
+    case toolNames.edit:
    case "Edit": {
-      const path = input?.abs_path ?? input?.file_path;
+      const path = input?.file_path ?? input?.file_path;
      let oldText = input.old_string ?? null;
      let newText = input.new_string ?? "";
      let affectedLines: number[] = [];
@@ -181,66 +188,13 @@ export function toolInfoFromToolUse(
      };
    }

-    case "mcp__acp__multi-edit":
-    case "MultiEdit": {
-      const multiInput = input as {
-        file_path: string;
-        edits: {
-          old_string: string;
-          new_string: string;
-          replace_all?: boolean;
-        }[];
-      };
-      let oldTextMulti = multiInput.edits.map((edit: any) => edit.old_string).join("\n");
-      let newTextMulti = multiInput.edits.map((edit: any) => edit.new_string).join("\n");
-      let affectedLines: number[] = [];
-      try {
-        if (multiInput.edits && multiInput.file_path) {
-          const oldContent =
-            cachedFileContent[multiInput.file_path] ||
-            multiInput.edits.map((edit: any) => edit.oldText).join("\n");
-          const newContent = replaceAndCalculateLocation(
-            oldContent,
-            multiInput.edits.map((edit) => ({
-              oldText: edit.old_string,
-              newText: edit.new_string,
-              replaceAll: edit.replace_all,
-            })),
-          );
-          oldTextMulti = oldContent;
-          newTextMulti = newContent.newContent;
-          affectedLines = newContent.lineNumbers;
-        }
-      } catch (e) {
-        console.error(e);
-      }
-      // Display it as a normal edit, because end users don't care about
-      // the distinction between edits and multi-edits.
-      return {
-        title: input?.file_path ? `Edit ${input.file_path}` : "Edit",
-        kind: "edit",
-        content: [
-          {
-            type: "diff" as const,
-            path: input.file_path,
-            oldText: oldTextMulti,
-            newText: newTextMulti,
-          },
-        ],
-        locations: input?.file_path
-          ? affectedLines.length > 0
-            ? affectedLines.map((line) => ({ line, path: input.file_path }))
-            : [{ path: input.file_path }]
-          : [],
-      };
-    }
-    case "mcp__acp__write": {
+    case toolNames.write: {
      let content: ToolCallContent[] = [];
-      if (input && input.abs_path) {
+      if (input && input.file_path) {
        content = [
          {
            type: "diff",
-            path: input.abs_path,
+            path: input.file_path,
            oldText: null,
            newText: input.content,
          },
@@ -254,10 +208,10 @@ export function toolInfoFromToolUse(
        ];
      }
      return {
-        title: input?.abs_path ? `Write ${input.abs_path}` : "Write",
+        title: input?.file_path ? `Write ${input.file_path}` : "Write",
        kind: "edit",
        content,
-        locations: input?.abs_path ? [{ path: input.abs_path }] : [],
+        locations: input?.file_path ? [{ path: input.file_path }] : [],
      };
    }

@@ -393,24 +347,30 @@ export function toolInfoFromToolUse(

    case "TodoWrite":
      return {
-        title: input?.todos
+        title: Array.isArray(input?.todos)
          ? `Update TODOs: ${input.todos.map((todo: any) => todo.content).join(", ")}`
          : "Update TODOs",
        kind: "think",
        content: [],
      };

-    case "exit_plan_mode":
+    case "ExitPlanMode":
      return {
-        title: "Exit Plan Mode",
-        kind: "think",
+        title: "Ready to code?",
+        kind: "switch_mode",
        content:
          input && input.plan
            ? [{ type: "content", content: { type: "text", text: input.plan } }]
            : [],
      };

-    case "Other":
+    case "Other": {
+      let output;
+      try {
+        output = JSON.stringify(input, null, 2);
+      } catch {
+        output = typeof input === "string" ? input : "{}";
+      }
      return {
        title: name || "Unknown Tool",
        kind: "other",
@@ -419,11 +379,12 @@ export function toolInfoFromToolUse(
            type: "content",
            content: {
              type: "text",
-              text: `\`\`\`json\n${JSON.stringify(input, null, 2) || "{}"}\`\`\``,
+              text: `\`\`\`json\n${output}\`\`\``,
            },
          },
        ],
      };
+    }

    default:
      return {
@@ -434,20 +395,21 @@ export function toolInfoFromToolUse(
  }
 }

-type ToolResult = {
-  type: "tool_result";
-  content: any;
-  tool_use_id: string;
-  is_error: boolean;
-};
-
 export function toolUpdateFromToolResult(
-  toolResult: ToolResult,
+  toolResult:
+    | ToolResultBlockParam
+    | BetaWebSearchToolResultBlockParam
+    | BetaWebFetchToolResultBlockParam
+    | WebSearchToolResultBlockParam
+    | BetaCodeExecutionToolResultBlockParam
+    | BetaBashCodeExecutionToolResultBlockParam
+    | BetaTextEditorCodeExecutionToolResultBlockParam
+    | BetaRequestMCPToolResultBlockParam,
  toolUse: any | undefined,
 ): ToolUpdate {
  switch (toolUse?.name) {
    case "Read":
-    case "mcp__acp__read":
+    case toolNames.read:
      if (Array.isArray(toolResult.content) && toolResult.content.length > 0) {
        return {
          content: toolResult.content.map((content: any) => ({
@@ -476,22 +438,28 @@ export function toolUpdateFromToolResult(
      }
      return {};

-    case "mcp__acp__Bash":
-    case "mcp__acp__edit":
+    case toolNames.bash:
    case "edit":
    case "Edit":
-    case "mcp__acp__multi-edit":
-    case "multi-edit":
-    case "MultiEdit":
-    case "mcp__acp__write":
+    case toolNames.edit:
+    case toolNames.write:
    case "Write": {
-      if (toolResult.is_error && toolResult.content?.length > 0) {
+      if (
+        "is_error" in toolResult &&
+        toolResult.is_error &&
+        toolResult.content &&
+        toolResult.content.length > 0
+      ) {
        // Only return errors
-        return toAcpContentUpdate(toolResult.content);
+        return toAcpContentUpdate(toolResult.content, true);
      }
      return {};
    }

+    case "ExitPlanMode": {
+      return { title: "Exited Plan Mode" };
+    }
+
    case "Task":
    case "NotebookEdit":
    case "NotebookRead":
@@ -507,17 +475,29 @@ export function toolUpdateFromToolResult(
    case "WebSearch":
    case "Other":
    default: {
-      return toAcpContentUpdate(toolResult.content);
+      return toAcpContentUpdate(
+        toolResult.content,
+        "is_error" in toolResult ? toolResult.is_error : false,
+      );
    }
  }
 }

-function toAcpContentUpdate(content: any): { content?: ToolCallContent[] } {
+function toAcpContentUpdate(
+  content: any,
+  isError: boolean = false,
+): { content?: ToolCallContent[] } {
  if (Array.isArray(content) && content.length > 0) {
    return {
      content: content.map((content: any) => ({
        type: "content",
-        content,
+        content:
+          isError && content.type === "text"
+            ? {
+                ...content,
+                text: `\`\`\`\n${content.text}\n\`\`\``,
+              }
+            : content,
      })),
    };
  } else if (typeof content === "string" && content.length > 0) {
@@ -527,7 +507,7 @@ function toAcpContentUpdate(content: any): { content?: ToolCallContent[] } {
          type: "content",
          content: {
            type: "text",
-            text: content,
+            text: isError ? `\`\`\`\n${content}\n\`\`\`` : content,
          },
        },
      ],
@@ -536,7 +516,7 @@ function toAcpContentUpdate(content: any): { content?: ToolCallContent[] } {
  return {};
 }

-type ClaudePlanEntry = {
+export type ClaudePlanEntry = {
  content: string;
  status: "pending" | "in_progress" | "completed";
  activeForm: string;
@@ -559,3 +539,50 @@ export function markdownEscape(text: string): string {
  }
  return escape + "\n" + text + (text.endsWith("\n") ? "" : "\n") + escape;
 }
+
+/* A global variable to store callbacks that should be executed when receiving hooks from Claude Code */
+const toolUseCallbacks: {
+  [toolUseId: string]: {
+    onPostToolUseHook?: (
+      toolUseID: string,
+      toolInput: unknown,
+      toolResponse: unknown,
+    ) => Promise<void>;
+  };
+} = {};
+
+/* Setup callbacks that will be called when receiving hooks from Claude Code */
+export const registerHookCallback = (
+  toolUseID: string,
+  {
+    onPostToolUseHook,
+  }: {
+    onPostToolUseHook?: (
+      toolUseID: string,
+      toolInput: unknown,
+      toolResponse: unknown,
+    ) => Promise<void>;
+  },
+) => {
+  toolUseCallbacks[toolUseID] = {
+    onPostToolUseHook,
+  };
+};
+
+/* A callback for Claude Code that is called when receiving a PostToolUse hook */
+export const postToolUseHook: HookCallback = async (
+  input: any,
+  toolUseID: string | undefined,
+): Promise<{ continue: boolean }> => {
+  if (input.hook_event_name === "PostToolUse" && toolUseID) {
+    const onPostToolUseHook = toolUseCallbacks[toolUseID]?.onPostToolUseHook;
+    if (onPostToolUseHook) {
+      await onPostToolUseHook(toolUseID, input.tool_input, input.tool_response);
+      delete toolUseCallbacks[toolUseID]; // Cleanup after execution
+    } else {
+      console.error(`No onPostToolUseHook found for tool use ID: ${toolUseID}`);
+      delete toolUseCallbacks[toolUseID];
+    }
+  }
+  return { continue: true };
+};
--- a/src/utils.ts
+++ b/src/utils.ts
@@ -2,6 +2,8 @@

 import { Readable, Writable } from "node:stream";
 import { WritableStream, ReadableStream } from "node:stream/web";
+import { readFileSync } from "node:fs";
+import { platform } from "node:os";

 // Useful for bridging push-based and async-iterator-based code.
 export class Pushable<T> implements AsyncIterable<T> {
@@ -74,9 +76,119 @@ export function nodeToWebReadable(nodeStream: Readable): ReadableStream<Uint8Arr
 }

 export function unreachable(value: never): never {
-  throw new Error(`Unexpected case: ${value}`);
+  let valueAsString;
+  try {
+    valueAsString = JSON.stringify(value);
+  } catch {
+    valueAsString = value;
+  }
+  throw new Error(`Unexpected case: ${valueAsString}`);
 }

 export function sleep(time: number): Promise<void> {
  return new Promise((resolve) => setTimeout(resolve, time));
 }
+
+interface ManagedSettings {
+  permissions?: {
+    allow?: string[];
+    deny?: string[];
+  };
+  env?: Record<string, string>;
+}
+
+// Following the rules in https://docs.anthropic.com/en/docs/claude-code/settings#settings-files
+// This can be removed once the SDK supports it natively.
+function getManagedSettingsPath(): string {
+  const os = platform();
+  switch (os) {
+    case "darwin":
+      return "/Library/Application Support/ClaudeCode/managed-settings.json";
+    case "linux": // including WSL
+      return "/etc/claude-code/managed-settings.json";
+    case "win32":
+      return "C:\\ProgramData\\ClaudeCode\\managed-settings.json";
+    default:
+      return "/etc/claude-code/managed-settings.json";
+  }
+}
+
+export function loadManagedSettings(): ManagedSettings | null {
+  try {
+    return JSON.parse(readFileSync(getManagedSettingsPath(), "utf8")) as ManagedSettings;
+  } catch {
+    return null;
+  }
+}
+
+export function applyEnvironmentSettings(settings: ManagedSettings): void {
+  if (settings.env) {
+    for (const [key, value] of Object.entries(settings.env)) {
+      process.env[key] = value;
+    }
+  }
+}
+
+export interface ExtractLinesResult {
+  content: string;
+  wasLimited: boolean;
+  linesRead: number;
+}
+
+/**
+ * Extracts lines from file content with byte limit enforcement.
+ *
+ * @param fullContent - The complete file content
+ * @param maxContentLength - Maximum number of UTF-16 Code Units to return
+ * @returns Object containing extracted content and metadata
+ */
+export function extractLinesWithByteLimit(
+  fullContent: string,
+  maxContentLength: number,
+): ExtractLinesResult {
+  if (fullContent === "") {
+    return {
+      content: "",
+      wasLimited: false,
+      linesRead: 1,
+    };
+  }
+
+  let linesSeen = 0;
+  let index = 0;
+  linesSeen = 0;
+
+  let contentLength = 0;
+  let wasLimited = false;
+
+  while (true) {
+    const nextIndex = fullContent.indexOf("\n", index);
+
+    if (nextIndex < 0) {
+      // Last line in file (no trailing newline)
+      if (linesSeen > 0 && fullContent.length > maxContentLength) {
+        wasLimited = true;
+        break;
+      }
+      linesSeen += 1;
+      contentLength = fullContent.length;
+      break;
+    } else {
+      // Line with newline - include up to the newline
+      const newContentLength = nextIndex + 1;
+      if (linesSeen > 0 && newContentLength > maxContentLength) {
+        wasLimited = true;
+        break;
+      }
+      linesSeen += 1;
+      contentLength = newContentLength;
+      index = newContentLength;
+    }
+  }
+
+  return {
+    content: fullContent.slice(0, contentLength),
+    wasLimited,
+    linesRead: linesSeen,
+  };
+}
Author	SHA1	Message	Date
Ben Brandt	6a0d0b5522	Fix incorrect stop reason after tool call rejection (#153 ) We were mapping tool call rejections to actual "refusals" which isn't quite right in ACP. Also filters out more of these status messages that Claude is returning.	2025-11-13 11:19:48 +01:00
Ben Brandt	7e0d2c9506	Update to claude-agent-sdk 0.1.37 (#152 )	2025-11-13 09:56:19 +00:00
Guillaume Sabran	2835f030cf	Add metadata to tool calls, including structured tool output (#128 ) Use the `_meta` field for tool calls to forward: - the original tool name (ie `Bash` etc) - the structured output of the tool call The output is received by using the `PostToolUse` hook from the SDK. The `tool_call_update` for the tool result is now only sent when both the hook callback and the original tool result (plain text) have been received. Claude Code describes [in its documentation ](https://docs.claude.com/en/api/agent-sdk/typescript#tool-output-types)the expected structure of the output for each tools. Unfortunately those are not part of the SDK, and [empirically I've found that there was some differences](`58c3100a36/local-server/src/server/endpoints/sendMessage/acp/clients/claudeCode/types.ts (L265)`). Still I've found that having this structured output was useful for my client to provide a richer UX. Maybe Zed will too. I was not sure how to add tests for this change, given the current setup	2025-11-13 10:42:21 +01:00
Ben Brandt	2710e146f4	Use vendored claude cli for experimental login method (#143 )	2025-11-05 11:36:02 +01:00
Richard Feldman	bf396fea67	Use terminal-auth from _meta if supported (#141 )	2025-11-04 21:34:16 -05:00
Ben Brandt	84b5744a2f	prep v0.10.0 (#140 )	2025-11-04 11:30:01 +00:00
Ben Brandt	b213152a94	Migrate from permission MCP server to canUseTool callback (#139 ) - Migrate from permission MCP server to canUseTool callback - Use SDK suggestions for permission updates	2025-11-04 12:26:25 +01:00
Ben Brandt	cd5aabdd74	Update to claude-agent-sdk 0.1.30 (#138 )	2025-11-04 10:38:20 +00:00
Bennet Bo Fenner	9e18a4e6fd	Add docs for release process (#135 )	2025-10-31 18:12:11 +01:00
Bennet Bo Fenner	48df285bd0	Prep v0.9.0 (#134 )	2025-10-31 16:50:55 +01:00
Bennet Bo Fenner	b138c84cdb	Support slash commands coming from MCP prompts (#133 ) Closes #115. Thanks to @joeljohansson99 for reporting/debugging.	2025-10-31 16:32:30 +01:00
Bennet Fenner	0735aca015	Update `UNSUPPORTED_COMMANDS` (#132 ) The SDK seems to now filter out slash commands that result in errors/crashes. Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-10-30 17:17:47 +01:00
Ben Brandt	847d62dacb	Revert entrypoint changes (#127 )	2025-10-27 12:45:58 +00:00
Ben Brandt	becf1d4f0a	Update to claude agent sdk 0.1.27 (#125 ) Also simplifies the permission mode workaround so we don't have to change the profile immediately, which could have raced with client changes anwyay.	2025-10-27 12:34:43 +01:00
Guillaume Sabran	5491bb8dcc	Allow for the package to be used as a dependency, while preserving existing behavior when run (#124 ) This change will allow for this package to be used directly, without launching a subprocess and communicating over stdio. --------- Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-10-27 11:10:41 +00:00
Ben Brandt	17aad5b7a1	Provide `agentInfo` on initialization response (#123 ) - Provide agentInfo on initialization - Prep 0.6.10	2025-10-24 17:02:07 +00:00
Steffen Deusch	dadc7f9f40	prevent invalid locations when read call has no path (#121 ) The ACP spec has the path as a required field for tool call locations, but I've seen Claude generate read calls without an input path, which would generate: ```json { "content": [], "kind": "read", "locations": [ { "line": 0 } ], "rawInput": {}, "sessionUpdate": "tool_call", "status": "pending", "title": "Read File", "toolCallId": "toolu_012PzeR5whUjWN4ij2HBubJw" } ``` And then fail to parse as SessionNotification in the SDK.	2025-10-24 12:49:43 +00:00
Ben Brandt	552f0a286d	Fix job name	2025-10-24 13:05:27 +02:00
Ben Brandt	589e1b5fc9	Update claude-agent-sdk to 0.1.26 (#120 )	2025-10-24 11:03:18 +00:00
Ben Brandt	0d97bb4753	Fix type holes in switch statements (#119 ) We were relying on unreachable to catch stuff for us, but it wasn't erroring because we were manually casting a few of the types to never, leading to a bug when a hook response was encountered. We'll revisit showing these hook responses to the user once ACP supports multiple types of status messages.	2025-10-24 10:19:41 +00:00
Ben Brandt	e3119b81f9	Prep v0.6.8 (#114 )	2025-10-21 09:19:37 +00:00
Ben Brandt	cd3c9f7b13	Update to claude-agent-sdk v0.1.23 (#113 ) - Don't emit content from message start - Update to claude-agent-sdk v0.1.23	2025-10-21 11:16:35 +02:00
Ben Brandt	8aea8fd0e1	Remove extra console.log	2025-10-17 13:35:36 +02:00
Ben Brandt	81a541c644	Fix for invalid plan input in the model (#110 ) Seems to have been introduced in the latest claude agent sdk. Closes #109	2025-10-17 11:33:46 +00:00
Ben Brandt	b49154bef8	Don't allow bypassPermissions to be enabled if in root/sudo mode, as Claude Code won't start (#108 ) Closes #104 Turns out this was the root cause, not the order of messages after all.	2025-10-17 10:58:38 +00:00
Ben Brandt	5fa453a33a	Reorganize order of loading config from the query. (#107 )	2025-10-17 12:36:58 +02:00
Ben Brandt	c2ab2da5ca	Remove duplicated text content after streaming (#106 )	2025-10-17 10:12:58 +00:00
Ben Brandt	373d940eb8	Update to claude-agent-sdk 0.1.21 and support streaming partial messages (#105 )	2025-10-17 11:58:36 +02:00
Ben Brandt	2c16a41fa4	prep 0.6.3 (#102 )	2025-10-16 12:35:47 +00:00
Ben Brandt	656cbe5921	Fix issue where slash commands were loaded before initialization was complete (#101 )	2025-10-16 14:34:17 +02:00
Ben Brandt	2f811f8a85	Update to claude-agent-sdk 0.1.19 (#100 )	2025-10-16 12:22:31 +00:00
Ben Brandt	6aa6537f67	Fix initialization error (#99 ) Closes #98 It turns out we can't always call this right away, as on some machines the process hasn't initialized yet. Since available models waits for the process to start, as a workaround we wait for that first, and then set the permission mode again once we know that it is ready.	2025-10-16 12:17:32 +00:00
Ben Brandt	1b72ac7c8a	Fix for allowing bypassPermissions to work (#97 ) It seems that unless you start the session with bypassPermissions, you can't change to it later. So we start with it on, and immediately change it back to default. This allows people to change it later however. Closes https://github.com/zed-industries/zed/issues/39392	2025-10-15 13:23:26 +00:00
Ben Brandt	6ae2c05ddb	Prep v0.6.0 (#96 )	2025-10-15 10:25:43 +00:00
Bennet Fenner	8f0fc59a48	Add model selection support (#93 ) This part of ACP is not yet stabilized, so marking as Draft for now --------- Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com> Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>	2025-10-15 12:21:30 +02:00
Ben Brandt	c25150523a	Update claude-agent-sdk to 0.1.15 (#95 ) - Update claude-agent-sdk to 0.1.15 - v0.5.6	2025-10-15 09:05:16 +00:00
WhiskeyJack96	ea796f3cbd	Add support for appending or overriding the system prompt (#91 ) resolves #90 Tested both the `append` and `override` flows by sending: ```javascript const resp = await this.connection.newSession({ cwd: basePath, _meta: { systemPrompt: "you are an obsidian assistant agent, you should be helpful to the user when working with their notes" }, mcpServers: [] }); ``` ```javascript const resp = await this.connection.newSession({ cwd: basePath, _meta: { systemPrompt: { append: "always speak in spanish" } }, mcpServers: [] }); ``` and confirmed that in both cases those prompts are what gets sent to the agent! --------- Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-10-15 10:48:06 +02:00
WhiskeyJack96	7f5ebca3fb	Make write and edit permissions consistent (#92 ) Currently edit requires a permissions call and write does not. This means that if clients want to allow users to approve `Write` calls they need to add handling to `HandleFileWrite` but if they do that then edits get _two_ permissions checks. Alternatively we could remove the write tool from approved tools (which would make more sense to me) but I'm assuming its there for a good reason, if not happy to change to remove it instead! --------- Co-authored-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-10-15 10:34:49 +02:00
Ben Brandt	75b44e945e	Update to claude-agent-sdk 0.1.13 (#89 )	2025-10-10 14:43:44 +00:00
Ben Brandt	ad8611c399	Migrate to @agentclientprotocol/sdk (#88 )	2025-10-10 16:17:37 +02:00
ozer	c5abff0e43	Improve setSessionMode error handling for SDK messages (#85 ) When `default`, `acceptEdits`, or `plan` modes fail, users see the real error from the SDK instead of a generic message. `Bypass permissions` gets special treatment - it shows both the SDK error AND explains it needs proper configuration. I couldn't test SDK thats why I added special treatment for bypass permissions. If you can test it we can talk further. Related to: https://github.com/zed-industries/zed/pull/39524 @benbrandt	2025-10-09 12:49:44 +02:00
Ben Brandt	5c98bd0404	Update to claude-agent-sdk@1.0.11 (#87 ) Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-10-09 08:30:56 +00:00
Alfredo Luque	71b660ae3b	Enable setting CLAUDE_CODE_EXECUTABLE to override the claude code CLI used (#86 ) ## Description Closes #84 by allowing users to set `CLAUDE_CODE_EXECUTABLE` to override the binary that's used by the claude code SDK. This is key in enterprise environments where claude CLI has been shimmed in some way.	2025-10-07 22:54:50 +02:00
Ben Brandt	4ee25af3f9	Update to claude-agent-sdk 0.1.8 (#83 )	2025-10-06 09:01:40 +00:00
Ben Brandt	1a5a9c16c6	Add back @anthropic-ai/claude-code as runtime dependency (#79 ) Some clients (i.e. Zed) still relied on this package being there at runtime. Adding it as a dependency for now.	2025-09-30 16:46:04 +02:00
Ben Brandt	5ca27369ef	Prep v0.5.1 (#78 )	2025-09-30 09:19:37 +00:00
Ben Brandt	8068d894c1	Align ACP tools with latest Claude Code tools (#77 ) Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-30 11:08:24 +02:00
Ben Brandt	71e745d5fb	Use default behavior for bypassPermissions mode (#75 ) They updated the types to allow us to just pass this through now. So we shouldn't have to have the special casing here.	2025-09-29 23:13:12 +02:00
Ben Brandt	2aa181be27	Migrate to Claude Agent SDK (#74 ) - Migrate to Claude Agent SDK - Prep v0.5	2025-09-29 21:58:38 +02:00
Ben Brandt	600a2db82f	More efficient file reads (#73 ) Allows the client to handle line offsets, rather than loading the entire file and then slicing it on the agent side. Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com> Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-29 13:54:30 +02:00
Ben Brandt	f2a4b3d19b	claude 1.0.128 (#72 ) - Update to 1.0.128 + use anthropic sdk types - Fix compacting output - Handle JSON errors --------- Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-29 12:59:59 +02:00
Ben Brandt	9135a0f1be	Update to claude-code-sdk v1.0.124 (#68 )	2025-09-25 15:05:47 +02:00
Ben Brandt	92c5cd84e8	Update clients list (#67 )	2025-09-25 12:26:40 +00:00
Christian Tietze	8311a7a222	add Emacs client to README (#66 )	2025-09-25 12:21:43 +00:00
Ben Brandt	ba513b190b	Claude Code SDK 1.0.123 (#65 ) - Bump to claude code sdk v1.0.123 - Prep release --------- Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-24 10:36:14 +02:00
Ben Brandt	a011363320	v0.4.3 (#63 )	2025-09-19 12:16:16 +02:00
Ben Brandt	7737878dd2	Pull permissions into own server (#50 ) This allows us to use the SDK version of an MCP server for all other tool calls. Ideally the permission server goes away eventually. Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com> --------- Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-19 12:05:46 +02:00
Ben Brandt	2e53f1d3e4	Update to ACP TS library v0.4 (#59 ) Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-18 10:04:30 +00:00
Ben Brandt	aa7364d49b	v0.4.2 (#55 )	2025-09-15 18:48:44 +02:00
Ben Brandt	e22d1ed52c	v0.4.1 (#54 ) Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-15 18:42:07 +02:00
Ben Brandt	ecfd36afa6	Support /compact (#51 ) It would be nice to also support sending status via hooks when the compact action is happening (and maybe show something in the UI to show where compacting happened in the thread) but for now this at least allows it. --------- Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-15 18:27:43 +02:00
Richard Feldman	7bd163818b	Add default limits to read tool (#52 ) Now it won't read files over 50K code units by default, and instead insist on breaking them up by line number.	2025-09-15 11:03:01 -04:00
Richard Feldman	491efe32e8	Tool error backticks (#53 ) This looks more reasonable now, because the code is now rendered in a monospace font: <img width="475" height="450" alt="Screenshot 2025-09-15 at 10 58 20 AM" src="https://github.com/user-attachments/assets/9870436b-ee8f-44cd-b814-b22ad9261ebe" /> That said, this definitely would benefit from a horizontal scroll bar on the code blocks. You can't really read what it's saying without copying the text out of the block. Of course, in this particular case the relevant info is all in the beginning of the error message (namely, that it failed to find the `old_string`), and this does look overall better than before (where it was just this big jumble of non-monospaced code), so I think it's worth merging as-is and then revisiting horizontal scrolling later.	2025-09-15 11:02:47 -04:00
Ben Brandt	ee17e45030	Use Trusted Publishing for npm release (#49 )	2025-09-15 10:59:16 +02:00
Ben Brandt	468c3522a0	Update deps to latest versions (#48 ) Signed-off-by: Ben Brandt <benjamin.j.brandt@gmail.com>	2025-09-15 10:46:05 +02:00
Ben Brandt	84d64b2762	Add build step to release	2025-09-15 10:35:21 +02:00
Myles Scolnick	9ff7919ea4	readme: Add marimo notebook (#47 )	2025-09-15 10:11:14 +02:00
Agus Zubiaga	d58c2dd45d	Update to acp 0.2.1 (#45 )	2025-09-12 14:07:00 -03:00
Richard Feldman	55532ecc57	Re-upgrade to SDK version 1.0.111 (#44 ) There's apparently a RCE on 1.0.100, so we shouldn't stay on that version.	2025-09-11 12:50:19 -04:00
Richard Feldman	f4711e96b0	Pin npm dependencies (#43 ) We discussed this and decided we'll manually upgrade on our own cadence. Don't want surprises like https://github.com/zed-industries/zed/issues/37976	2025-09-11 12:27:52 -04:00
Richard Feldman	f691024350	Load managed-settings.json (#42 ) The SDK doesn't currently load [Claude Code's `managed-settings.json`](https://docs.anthropic.com/en/docs/claude-code/settings#settings-files) - now the adapter does (following [the rules in the docs](https://docs.anthropic.com/en/docs/claude-code/settings#settings-files)), but we can remove this once the SDK supports it.	2025-09-11 08:47:33 -04:00
Agus Zubiaga	d6c91b1ec1	0.4.0	2025-09-09 11:49:37 -03:00
Agus Zubiaga	0289dd9de7	Support plan mode and all other permission modes (#40 ) Exposes all Claude Code permission modes as [ACP Session modes](https://github.com/zed-industries/agent-client-protocol/pull/67). This allows ACP clients to build mode selectors and to be notified when the mode changes as a result of `ExitPlanMode`. See a demo in the [Zed PR](https://github.com/zed-industries/zed/pull/37632). Closes #31 --------- Co-authored-by: Bennet Bo Fenner <bennetbo@gmx.de> Co-authored-by: Richard Feldman <oss@rtfeldman.com> Co-authored-by: Bennet <bennet@zed.dev>	2025-09-09 11:48:26 -03:00
Agus Zubiaga	5cd3ac91e8	Removing "Coming soon"	2025-09-08 13:18:28 -03:00
Agus Zubiaga	6b5873f005	0.3.0	2025-09-08 13:13:25 -03:00
Agus Zubiaga	539f94de45	Support HTTP/SSE MCP servers (#38 ) HTTP/SSE MCP servers can be now included in `session/new`	2025-09-08 13:12:19 -03:00
Jie Wang	70441b0aa3	Move `@types/express` to dev dependencies (#34 )	2025-09-05 06:19:09 +02:00
Agus Zubiaga	de514862f8	0.2.6	2025-09-04 18:01:30 -03:00
Ben Brandt	9ebdfffd48	Remove option for using an API Key, just support /login Co-authored-by: Agus Zubiaga <agus@zed.dev>	2025-09-04 08:41:45 -07:00