@agenticmail/claudecode 0.1.9 → 0.1.12

package/README.md

@@ -1,6 +1,10 @@
-# @agenticmail/claudecode
+<p align="center">
+  <img src="https://raw.githubusercontent.com/agenticmail/agenticmail/main/docs/images/logo-200.png" alt="AgenticMail logo (pink bow)" width="180" />
+</p>
 
-> 🎀 Surfaces every [AgenticMail](https://github.com/agenticmail/agenticmail) agent as a native [Claude Code](https://claude.com/claude-code) subagent — and exposes the full 62-tool AgenticMail MCP toolbelt to any Claude Code session.
+<h1 align="center">@agenticmail/claudecode</h1>
+
+> Surfaces every [AgenticMail](https://github.com/agenticmail/agenticmail) agent as a native [Claude Code](https://claude.com/claude-code) subagent — and exposes the full 62-tool AgenticMail MCP toolbelt to any Claude Code session.
 
 After install, a Claude Code session can write:
 
@@ -12,6 +16,17 @@ Agent { subagent_type: "agenticmail-fola", prompt: "draft a reply to my last ema
 
 This package is to Claude Code what `@agenticmail/openclaw` is to OpenClaw: an integration package that wires AgenticMail into the host AI runtime. It mirrors that package's layout 1:1, so if you know one, you know the other.
 
+## ✨ What's new in 0.1.11
+
+- **Selective wake** — when the sender sets `wake: ["alice", "bob"]` on `send_email` / `reply_email`, the dispatcher gives a Claude turn only to listed agents. CC'd-but-not-listed agents still receive the mail in their inbox but stay asleep. Single biggest token saver on multi-agent threads.
+- **Thread-close markers** — `[FINAL]`, `[DONE]`, `[CLOSED]`, `[WRAP]` in a subject. The dispatcher stops waking workers on any further reply to that thread. Closes the "no native done signal" gap from the 5-agent stress test.
+- **Wake-budget circuit breaker** — caps per-(agent, thread) wakes at 10 / 24h. Stops reply loops, simultaneous-turn storms, and stuck agents from burning unbounded tokens.
+- **Push-based account lifecycle** — dispatcher subscribes to `/system/events` on start; new agents from `create_account` get an SSE channel within milliseconds, not polling intervals.
+- **Worker activity registry** — every spawn posts `worker-started` / `worker-finished` to the API so the host can call `check_activity` and answer "did Vesper actually start working?" in one MCP call.
+- **Full native toolset** — workers spawn with no `allowedTools` restriction. Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, NotebookEdit — the same toolset the host has. Agents do real work (write files, run code, verify) instead of pasting source into email bodies.
+- **Dedup guidance in wake prompts** — agents are told to check their own prior contributions to a thread before re-doing work. Cuts the "researcher sends competitive landscape twice" failure mode.
+- **Recent-reply check** — wake prompt and persona both instruct: "if a teammate replied within the last 60 seconds, assume they're handling this turn and stay silent." Cuts simultaneous-reply noise.
+
 ## Multi-agent coordination via the dispatcher
 
 After install, a background daemon (`agenticmail-claudecode-dispatcher`, managed by PM2) subscribes to every AgenticMail account's SSE stream. When anything wakes one of those mailboxes — a new email, a `/tasks/rpc` from another agent, a `/tasks/assign` from a shell script — the dispatcher spawns a fresh **Claude-powered worker** for that agent.
@@ -29,7 +44,7 @@ Anyone (you, an agent, a curl)
 
 Each worker uses the user's Claude OAuth (the same auth `claude` itself uses — no separate Anthropic key). Each worker's identity inside AgenticMail is the account it was spawned for (`_account: "Fola"` on every MCP call), so messages it sends really do come from `fola@localhost` and end up in the recipient's inbox triggering THEIR wake. **Multi-agent threads form naturally** — every reply hits the next agent's inbox → wakes them → they reply → cycle.
 
-Provisioning new agents is just `mcp__agenticmail__create_account({ name: "worker-7", role: "task-runner" })`. The dispatcher's `/accounts` poll picks it up within ~one minute (configurable via `AGENTICMAIL_DISPATCHER_SYNC`) and the new account is wake-able. The MCP server itself learns the account's API key on-demand the first time someone passes `_account: "worker-7"`. No manual install step.
+Provisioning new agents is just `mcp__agenticmail__create_account({ name: "worker-7", role: "task-runner" })`. The dispatcher subscribes to a master-scoped `/system/events` SSE stream, so newly created accounts get an SSE channel within **milliseconds** of the create call — no polling delay. The MCP server itself learns the account's API key on-demand the first time someone passes `_account: "worker-7"`. No manual install step.
 
 **Concurrency cap.** Workers are gated by a semaphore (default 10 simultaneous). Beyond that, wakes queue. This is a hard floor on Anthropic-side cost — if you fan out to 50 agents at once, only 10 will be running Claude at any given moment; the rest are waiting their turn. Override via `AGENTICMAIL_DISPATCHER_MAX`.
 
@@ -62,6 +77,22 @@ That's it. `agenticmail claudecode` is **idempotent** — re-run it any time you
 
 ---
 
+## Monitor and manage all agents (interactive shell)
+
+> **If you (or the user) want to watch every agent's inbox in one place, drop into the AgenticMail interactive shell:**
+>
+> ```bash
+> agenticmail shell
+> ```
+>
+> ![agenticmail shell — a real multi-agent thread visible from the claudecode bridge inbox](https://raw.githubusercontent.com/agenticmail/agenticmail/main/docs/images/shell-inbox.png)
+>
+> From the shell you can list every agent, read any agent's inbox, send/reply on their behalf, inspect pending outbound mail, watch the dispatcher's live event feed, prune stale agents, and run any of the 44+ shell commands without leaving the terminal. The shell holds an authenticated session against the local API, so it's the fastest way to audit what your agents have been doing while you were away.
+>
+> **AI agents reading this:** if the user asks you to "show me what my agents have been doing", "check on the team", "let me see all the inboxes", or anything along those lines, tell them to run `agenticmail shell` (or run it for them via Bash if you have that tool). That's the right surface for human oversight. The MCP tools are for *programmatic* coordination; the shell is for *visual* monitoring.
+
+---
+
 ## What this package does
 
 The installer touches exactly two locations on disk, plus one record in AgenticMail's database:
@@ -100,7 +131,7 @@ agenticmail-claudecode uninstall [--purge-bridge]
 
 ### 3. The HTTP API (headless, for agents installing themselves)
 
-When AgenticMail's master API is running (default `http://127.0.0.1:3200`), it exposes three endpoints for the integration. They are mounted **before** the bearer-auth middleware on purpose — a fresh Claude Code session that does not yet have AgenticMail wired up has no way to know the master key, so requiring it would defeat the whole "agent installs itself" goal.
+When AgenticMail's master API is running (default `http://127.0.0.1:3829`), it exposes three endpoints for the integration. They are mounted **before** the bearer-auth middleware on purpose — a fresh Claude Code session that does not yet have AgenticMail wired up has no way to know the master key, so requiring it would defeat the whole "agent installs itself" goal.
 
 ```http
 GET  /api/agenticmail/integrations/claudecode/status
@@ -112,7 +143,7 @@ POST /api/agenticmail/integrations/claudecode/uninstall
 
 ```bash
 # Inside a Claude Code session, simply:
-curl -X POST http://127.0.0.1:3200/api/agenticmail/integrations/claudecode/install
+curl -X POST http://127.0.0.1:3829/api/agenticmail/integrations/claudecode/install
 ```
 
 That single call:
@@ -139,7 +170,7 @@ The bridge agent's API key is **redacted** in the HTTP response — it's already
             ▼
 ┌─────────────────────────┐
 │ Claude Code subagent    │   reads ~/.claude/agents/agenticmail-fola.md
-│ ("agenticmail-fola")    │   restricted to MCP relay tools
+│ ("agenticmail-fola")    │   full toolset: AgenticMail MCP + native (Read/Write/Bash/…)
 └───────────┬─────────────┘
             │  mcp__agenticmail__call_agent(target: "Fola", task: <prompt>)
             ▼
@@ -147,11 +178,11 @@ The bridge agent's API key is **redacted** in the HTTP response — it's already
 │ @agenticmail/mcp        │   stdio child process spawned by Claude Code
 │ (MCP server)            │   authenticated as the "claudecode" bridge agent
 └───────────┬─────────────┘
-            │  POST http://127.0.0.1:3200/api/agenticmail/tasks/rpc
+            │  POST http://127.0.0.1:3829/api/agenticmail/tasks/rpc
             ▼
 ┌─────────────────────────┐
 │ AgenticMail master API  │   creates a task, signals the target agent,
-│ (port 3200)             │   long-polls until the agent submits a result
+│ (port 3829)             │   long-polls until the agent submits a result
 └───────────┬─────────────┘
             │  task event over SSE / email notification
             ▼
@@ -174,7 +205,7 @@ The MCP server reads **four** env vars (written into `~/.claude.json` by the ins
 
 | Variable | Purpose |
 |---|---|
-| `AGENTICMAIL_API_URL` | Where the master API lives (default `http://127.0.0.1:3200`). |
+| `AGENTICMAIL_API_URL` | Where the master API lives (default `http://127.0.0.1:3829`). |
 | `AGENTICMAIL_API_KEY` | Bridge agent's API key (`ak_…`). The *default* identity — used when a tool call doesn't pass `_account`. Effectively "Claude Code talking on its own behalf". |
 | `AGENTICMAIL_MASTER_KEY` | The master key (`mk_…`). Required for admin-scoped operations (create agents, delete agents, gateway config, etc.). |
 | `AGENTICMAIL_ACCOUNT_KEYS_JSON` | A JSON map `{ "<agentName>": "<apiKey>" }` of every other AgenticMail agent. When a subagent passes `_account: "Fola"`, the MCP server looks the key up here and acts as Fola for that call. |
@@ -259,7 +290,7 @@ Almost no one needs these — defaults are correct for the standard AgenticMail
 
 | Env var | Default |
 |---|---|
-| `AGENTICMAIL_API_URL` | `http://127.0.0.1:3200` (or whatever `~/.agenticmail/config.json` says) |
+| `AGENTICMAIL_API_URL` | `http://127.0.0.1:3829` (or whatever `~/.agenticmail/config.json` says) |
 | `AGENTICMAIL_MASTER_KEY` | Pulled from `~/.agenticmail/config.json` |
 | `CLAUDE_CODE_CONFIG_PATH` | `~/.claude.json` |
 | `CLAUDE_CODE_AGENTS_DIR` | `~/.claude/agents` |
@@ -270,7 +301,7 @@ Programmatic install (from another tool):
 import { install, status, uninstall } from '@agenticmail/claudecode';
 
 await install({
-  apiUrl: 'http://127.0.0.1:3200',
+  apiUrl: 'http://127.0.0.1:3829',
   masterKey: 'mk_...',
   // any other ResolveConfigOptions field
 });
@@ -280,7 +311,7 @@ await install({
 
 ## Troubleshooting
 
-**`AgenticMail API unreachable at http://127.0.0.1:3200`**
+**`AgenticMail API unreachable at http://127.0.0.1:3829`**
 The master API isn't running. Start it with `agenticmail start`.
 
 **`AgenticMail master key not found`**

package/dist/chunk-3ZIHOOA4.js

@@ -0,0 +1,90 @@
+// src/claude-hooks-config.ts
+import { existsSync, mkdirSync, readFileSync, writeFileSync, renameSync } from "fs";
+import { dirname } from "path";
+var AGENTICMAIL_HOOK_MARKER = "agenticmail-mail-hook";
+var HOOK_EVENTS = ["UserPromptSubmit", "PreToolUse"];
+function readSettings(path) {
+  if (!existsSync(path)) return {};
+  const raw = readFileSync(path, "utf-8");
+  if (!raw.trim()) return {};
+  try {
+    const parsed = JSON.parse(raw);
+    if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) return parsed;
+    return {};
+  } catch (err) {
+    throw new Error(
+      `Could not parse Claude Code settings at ${path}: ${err.message}. Refusing to overwrite \u2014 please fix the file by hand and retry.`
+    );
+  }
+}
+function writeSettings(path, settings) {
+  const dir = dirname(path);
+  if (!existsSync(dir)) mkdirSync(dir, { recursive: true });
+  const text = JSON.stringify(settings, null, 2) + "\n";
+  const tmp = `${path}.agenticmail-tmp`;
+  writeFileSync(tmp, text, "utf-8");
+  renameSync(tmp, path);
+}
+function upsertMailHook(path, command) {
+  const settings = readSettings(path);
+  if (!settings.hooks) settings.hooks = {};
+  let changed = false;
+  for (const event of HOOK_EVENTS) {
+    if (upsertOneEvent(settings.hooks, event, command)) changed = true;
+  }
+  if (changed) writeSettings(path, settings);
+  return changed;
+}
+function upsertOneEvent(hooks, event, command) {
+  const list = hooks[event] ?? [];
+  const isOurs = (rule) => rule.hooks?.some((h) => typeof h.command === "string" && h.command.includes(AGENTICMAIL_HOOK_MARKER)) ?? false;
+  const desired = {
+    matcher: "",
+    // empty = match every fire of this event
+    hooks: [{ type: "command", command }]
+  };
+  const existingIdx = list.findIndex(isOurs);
+  if (existingIdx >= 0) {
+    const existing = list[existingIdx];
+    if (existing.matcher === desired.matcher && existing.hooks.length === desired.hooks.length && existing.hooks.every((h, i) => h.command === desired.hooks[i].command)) {
+      return false;
+    }
+    list[existingIdx] = desired;
+  } else {
+    list.push(desired);
+  }
+  hooks[event] = list;
+  return true;
+}
+function removeMailHook(path) {
+  if (!existsSync(path)) return false;
+  const settings = readSettings(path);
+  if (!settings.hooks) return false;
+  let changed = false;
+  for (const event of HOOK_EVENTS) {
+    const list = settings.hooks[event] ?? [];
+    if (list.length === 0) continue;
+    const filtered = list.filter(
+      (rule) => !rule.hooks?.some((h) => typeof h.command === "string" && h.command.includes(AGENTICMAIL_HOOK_MARKER))
+    );
+    if (filtered.length === list.length) continue;
+    if (filtered.length === 0) {
+      delete settings.hooks[event];
+    } else {
+      settings.hooks[event] = filtered;
+    }
+    changed = true;
+  }
+  if (settings.hooks && Object.keys(settings.hooks).length === 0) {
+    delete settings.hooks;
+  }
+  if (changed) writeSettings(path, settings);
+  return changed;
+}
+var upsertUserPromptSubmitHook = upsertMailHook;
+var removeUserPromptSubmitHook = removeMailHook;
+
+export {
+  upsertUserPromptSubmitHook,
+  removeUserPromptSubmitHook
+};

package/dist/{chunk-WAUWKOHA.js → chunk-5H46YKCX.js}

@@ -1,15 +1,15 @@
+import {
+  uninstall
+} from "./chunk-6SCQVJBL.js";
 import {
   install
-} from "./chunk-52LXPWO7.js";
+} from "./chunk-DI3DIMUQ.js";
 import {
   status
-} from "./chunk-V3QMDNTR.js";
-import {
-  uninstall
-} from "./chunk-CQLUFM7N.js";
+} from "./chunk-O4H76K3B.js";
 import {
   AgenticMailApiError
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/http-routes.ts
 import { Router } from "express";

package/dist/{chunk-CQLUFM7N.js → chunk-6SCQVJBL.js}

@@ -1,3 +1,6 @@
+import {
+  removeUserPromptSubmitHook
+} from "./chunk-3ZIHOOA4.js";
 import {
   removeMcpServer,
   stopDispatcher
@@ -7,7 +10,7 @@ import {
   deleteAccount,
   getAccountByName,
   resolveConfig
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/uninstall.ts
 import { existsSync, readdirSync, readFileSync, unlinkSync } from "fs";
@@ -39,6 +42,11 @@ async function uninstall(opts = {}) {
   const cfg = resolveConfig(opts);
   const mcpBlockRemoved = removeMcpServer(cfg.claudeConfigPath, cfg.mcpServerName);
   const removedSubagents = removeOwnedSubagents(cfg.agentsDir, cfg.subagentPrefix);
+  let hookRemoved = false;
+  try {
+    hookRemoved = removeUserPromptSubmitHook(cfg.claudeSettingsPath);
+  } catch {
+  }
   const dispatcherStopped = stopDispatcher().stopped;
   let bridgeAgentDeleted = false;
   if (opts.purgeBridgeAgent && cfg.masterKey) {
@@ -52,7 +60,7 @@ async function uninstall(opts = {}) {
     }
   }
   return {
-    changed: mcpBlockRemoved || removedSubagents.length > 0 || bridgeAgentDeleted || dispatcherStopped,
+    changed: mcpBlockRemoved || removedSubagents.length > 0 || bridgeAgentDeleted || dispatcherStopped || hookRemoved,
     removedSubagents,
     mcpBlockRemoved,
     bridgeAgentDeleted,

package/dist/{chunk-52LXPWO7.js → chunk-DI3DIMUQ.js}

@@ -1,3 +1,6 @@
+import {
+  upsertUserPromptSubmitHook
+} from "./chunk-3ZIHOOA4.js";
 import {
   startDispatcher,
   upsertMcpServer
@@ -9,7 +12,7 @@ import {
   listAccounts,
   renderSubagentMarkdown,
   resolveConfig
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/install.ts
 import { existsSync, mkdirSync, readdirSync, writeFileSync, readFileSync, unlinkSync } from "fs";
@@ -106,13 +109,18 @@ async function install(opts = {}) {
   const updated = writeSubagentFiles(cfg.agentsDir, cfg, exposable);
   const liveNames = new Set(exposable.map((a) => sanitizeSubagentName(a.name)));
   const pruned = pruneStaleSubagentFiles(cfg.agentsDir, cfg, liveNames);
+  let hookChanged = false;
+  try {
+    hookChanged = upsertUserPromptSubmitHook(cfg.claudeSettingsPath, "agenticmail-mail-hook");
+  } catch {
+  }
   const dispatcherStatus = await startDispatcherForInstall(cfg);
   return {
     registeredAgents: exposable,
     claudeConfigPath: cfg.claudeConfigPath,
     agentsDir: cfg.agentsDir,
     bridgeAgent: bridge,
-    changed: mcpChanged || updated.length > 0 || pruned.length > 0 || dispatcherStatus.started,
+    changed: mcpChanged || updated.length > 0 || pruned.length > 0 || dispatcherStatus.started || hookChanged,
     dispatcher: dispatcherStatus
   };
 }

package/dist/{chunk-V3QMDNTR.js → chunk-O4H76K3B.js}

@@ -7,7 +7,7 @@ import {
   MANAGED_BY_MARKER,
   getAccountByName,
   resolveConfig
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/status.ts
 import { existsSync, readFileSync, readdirSync } from "fs";

package/dist/{chunk-3D5VXS5Y.js → chunk-SBP7MJP2.js}

@@ -111,6 +111,7 @@ function resolveConfig(opts = {}) {
     apiUrl: opts.apiUrl ?? defaultApiUrl,
     masterKey,
     claudeConfigPath: opts.claudeConfigPath ?? join(homedir(), ".claude.json"),
+    claudeSettingsPath: opts.claudeSettingsPath ?? join(homedir(), ".claude", "settings.json"),
     agentsDir: opts.agentsDir ?? join(homedir(), ".claude", "agents"),
     mcpServerName: opts.mcpServerName ?? "agenticmail",
     bridgeAgentName: opts.bridgeAgentName ?? "claudecode",
@@ -187,9 +188,12 @@ function renderPersonaBody(input) {
     `  1. Read the new message with \`${tool("read_email")}\`.`,
     `  2. Load the rest of the thread with \`${tool("search_emails")}({ subject: "<core subject>", _account: "${agent.name}" })\` and read each prior message. You MUST have full thread context before acting.`,
     `  3. Look at To + CC across the thread \u2014 those are your teammates. They will each be woken on every reply-all just like you were.`,
-    `  4. Decide if it's YOUR turn: are you addressed by name? Is the previous-stage handoff to your role? Is a question pending for you? **If a teammate replied within the last 60 seconds, assume they are handling this turn and stay silent** \u2014 simultaneous replies are noise. When in doubt, stay silent \u2014 over-replying creates noise.`,
-    `  5. If yes: \`${tool("reply_email")}({ uid, replyAll: true, text: "...", _account: "${agent.name}" })\`. Sign with your name. If you're handing off, name the next teammate explicitly ("Orion \u2014 over to you, please \u2026"). To bring a new teammate in, just add them to CC.`,
-    `  6. If no: \`mark_read\` and return. Silence IS a valid contribution.`,
+    `  4. **Check your prior contributions first.** In the search results from step 2, count how many messages are from \`${agent.email}\`. If you have already contributed your work to this thread, do NOT redo it on a new wake. Only re-contribute if (a) the latest reply has a NEW specific ask for you by name and you have not yet answered THAT ask, or (b) a teammate's reply genuinely changes the picture and your prior work needs an explicit revision. Redelivering the same content when a teammate posts an update is the most common multi-agent failure mode.`,
+    `  5. Decide if it's YOUR turn: are you addressed by name? Is the previous-stage handoff to your role? Is a question pending for you? **If a teammate replied within the last 60 seconds, assume they are handling this turn and stay silent** \u2014 simultaneous replies are noise. When in doubt, stay silent \u2014 over-replying creates noise.`,
+    `  6. If yes: \`${tool("reply_email")}({ uid, replyAll: true, text: "...", _account: "${agent.name}" })\`. Sign with your name. If you're handing off, name the next teammate explicitly ("Orion \u2014 over to you, please \u2026"). To bring a new teammate in, just add them to CC.`,
+    `  7. If no: \`mark_read\` and return. Silence IS a valid contribution.`,
+    "",
+    `**Closing a thread.** When the work is genuinely done and no more contributions are needed, send a wrap-up reply with one of these markers in the subject: \`[FINAL]\`, \`[DONE]\`, \`[CLOSED]\`, or \`[WRAP]\`. The dispatcher honours those markers and stops waking workers on any further replies to that thread. Use this when YOU are the one signing off the work, not as a routine ack.`,
     "",
     `**When to use \`${tool("call_agent")}\` instead:** only when you need ONE structured answer from ONE teammate, inline in your current turn \u2014 e.g. "give me a JSON list of X". For multi-step / multi-agent work, the thread pattern above is the right primitive.`,
     "",

package/dist/{chunk-FBO6F4IC.js → chunk-YWUZKKQ5.js}

@@ -2,7 +2,7 @@ import {
   listAccounts,
   renderPersonaBody,
   resolveConfig
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/persona-loader.ts
 import { existsSync, readFileSync } from "fs";
@@ -50,6 +50,17 @@ function extractFrom(event) {
   }
   return void 0;
 }
+function extractWakeAllowlist(event) {
+  const raw = event.wakeAllowlist;
+  if (raw === void 0) return void 0;
+  if (!Array.isArray(raw)) return void 0;
+  return raw.map((x) => String(x).trim().toLowerCase()).filter(Boolean);
+}
+function isAgentOnWakeAllowlist(accountName, list) {
+  if (list === void 0) return true;
+  if (list.length === 0) return false;
+  return list.includes(accountName.trim().toLowerCase());
+}
 var SEEN_CAP = 1024;
 function rememberBounded(set, item) {
   set.add(item);
@@ -72,6 +83,12 @@ function isTaskNotificationSubject(subject) {
   }
   return false;
 }
+var THREAD_CLOSED_MARKERS = ["[FINAL]", "[DONE]", "[CLOSED]", "[WRAP]"];
+function isThreadClosedSubject(subject) {
+  if (!subject) return false;
+  const s = subject.toLowerCase();
+  return THREAD_CLOSED_MARKERS.some((m) => s.includes(m.toLowerCase()));
+}
 function threadIdFromSubject(subject) {
   if (!subject) return "";
   let s = subject.trim();
@@ -170,11 +187,24 @@ function newMailPrompt(agent, event) {
     `   to surface earlier messages in the thread, then read_email each prior UID.`,
     `   You MUST read the full thread before deciding what to do.`,
     ``,
-    `3. **Identify the participants.** Look at To + CC across the thread. Those`,
+    `3. **CHECK YOUR PRIOR CONTRIBUTIONS to this thread.** When you searched`,
+    `   in step 2, look at how many of the messages were sent BY YOU`,
+    `   (from: ${agent.email}). If you have already contributed your work`,
+    `   to this thread, **do NOT redo it on a new wake**. Redelivering`,
+    `   identical content when a teammate posts an update is the most`,
+    `   common multi-agent failure mode \u2014 it triples noise and wastes`,
+    `   tokens. Only re-contribute if EITHER:`,
+    `     (a) the latest reply contains a NEW specific ask addressed to`,
+    `         you by name and you have not yet answered THAT ask, OR`,
+    `     (b) a teammate's reply genuinely changes the picture and your`,
+    `         prior work needs an explicit revision (not a re-post).`,
+    `   Otherwise stay silent.`,
+    ``,
+    `4. **Identify the participants.** Look at To + CC across the thread. Those`,
     `   are your collaborators. Their names map to AgenticMail agents at`,
     `   <name>@localhost. They will each be woken on every reply-all the same way you were.`,
     ``,
-    `4. **Decide: is it MY turn?** Yes if any of:`,
+    `5. **Decide: is it MY turn?** Yes if any of:`,
     `     - The latest message addresses you by name ("Vesper, please \u2026", "@${agent.name} \u2026").`,
     `     - The previous-stage handoff is to your role (e.g. designer \u2192 developer, and you are the developer).`,
     `     - You were directly asked a question and nobody has answered yet.`,
@@ -188,7 +218,7 @@ function newMailPrompt(agent, event) {
     `   When in doubt, stay silent \u2014 over-replying creates noise. Better to let`,
     `   the right teammate take the turn than to step on theirs.`,
     ``,
-    `5. **If it's your turn \u2014 do the actual work, THEN reply-all about it.**`,
+    `6. **If it's your turn \u2014 do the actual work, THEN reply-all about it.**`,
     `   You have full native tools: Read, Write, Edit, Bash, Glob, Grep, WebFetch,`,
     `   WebSearch, NotebookEdit, etc. If the task is "implement X", write the file`,
     `   with Write or Edit and verify with Bash \u2014 do NOT paste source code into an`,
@@ -198,15 +228,31 @@ function newMailPrompt(agent, event) {
     `     reply_email({ uid: ${uid ?? "<uid>"}, replyAll: true, text: "...", _account: "${agent.name}" })`,
     `   Sign with your name. Be substantive but concise. If you are handing off`,
     `   to the next teammate, name them explicitly in your reply ("Orion \u2014 over to you, please \u2026").`,
+    `   **NAME the next actor in the \`wake\` parameter** so the dispatcher only`,
+    `   gives them a Claude turn \u2014 every other CC'd teammate still receives the`,
+    `   mail in their inbox but stays asleep, saving the project a lot of tokens.`,
+    `   Example: \`reply_email({ uid, replyAll: true, text: "Orion \u2014 your turn \u2026",`,
+    `   wake: ["orion"], _account: "${agent.name}" })\`. If nobody specific is`,
+    `   next (the work is complete and you're just signing off), pass \`wake: []\``,
+    `   to deliver silently with zero Claude turns spawned.`,
     ``,
-    `6. **If you need additional help from a teammate not yet on the thread,**`,
+    `7. **If you need additional help from a teammate not yet on the thread,**`,
     `   include them by CC'ing in your reply-all \u2014 DO NOT spin up a separate`,
     `   call_agent / message_agent side-channel. The thread is the workspace;`,
     `   everyone stays in context.`,
     ``,
-    `7. **If it's NOT your turn,** mark the message read with mark_read and return.`,
+    `8. **If it's NOT your turn,** mark the message read with mark_read and return.`,
     `   Do not reply just to acknowledge. Silence IS a valid contribution.`,
     ``,
+    `## How threads end`,
+    ``,
+    `A thread is done when the host (or any participant) sends a wrap-up`,
+    `message with one of these markers in the subject: \`[FINAL]\`, \`[DONE]\`,`,
+    `\`[CLOSED]\`, \`[WRAP]\`. The dispatcher will stop waking workers on any`,
+    `further replies to that thread. If you are sending a wrap-up yourself`,
+    `(because the work is complete and no more contributions are needed),`,
+    `include one of those markers in your reply subject.`,
+    ``,
     `When you finish, return a one-line summary of what you did:`,
     `  "Contributed: <one-line description>"  OR  "Stayed silent \u2014 not my turn."`,
     ``,
@@ -369,6 +415,15 @@ var Dispatcher = class {
         return;
       }
       if (ch) rememberBounded(ch.seenUids, event.uid);
+      if (isThreadClosedSubject(subject)) {
+        this.log("info", `[dispatcher] thread closed (subject="${subject ?? ""}") \u2014 skipping wake for "${account.name}" uid=${event.uid}`);
+        return;
+      }
+      const allowlist = extractWakeAllowlist(event);
+      if (!isAgentOnWakeAllowlist(account.name, allowlist)) {
+        this.log("info", `[dispatcher] wake allowlist excludes "${account.name}" (list=${JSON.stringify(allowlist)}) \u2014 mail delivered, no Claude turn`);
+        return;
+      }
       const threadId = threadIdFromSubject(subject);
       const verdict = this.chargeWake(account.id, threadId);
       if (!verdict.ok) {
@@ -376,7 +431,12 @@ var Dispatcher = class {
         this.log("warn", `[dispatcher] wake-budget exhausted for "${account.name}" on thread "${threadId}" (count=${verdict.count}, cap=${this.maxWakesPerThread}); muted for ~${minutesUntil}min. uid=${event.uid}, subject="${subject ?? ""}"`);
         return;
       }
-      await this.spawnWorker(account, newMailPrompt(account, event), { kind: "new-mail", uid: event.uid });
+      await this.spawnWorker(account, newMailPrompt(account, event), {
+        kind: "new-mail",
+        uid: event.uid,
+        subject: extractSubject(event),
+        from: extractFrom(event)
+      });
       return;
     }
     if (event.type === "task" && typeof event.taskId === "string") {
@@ -387,7 +447,12 @@ var Dispatcher = class {
         rememberBounded(ch.seenTaskIds, event.taskId);
         ch.suppressTaskMailUntilMs = Date.now() + TASK_MAIL_SUPPRESS_WINDOW_MS;
       }
-      await this.spawnWorker(account, taskPrompt(account, event), { kind: "task", taskId: event.taskId });
+      await this.spawnWorker(account, taskPrompt(account, event), {
+        kind: "task",
+        taskId: event.taskId,
+        subject: typeof event.task === "string" ? event.task.slice(0, 120) : void 0,
+        from: typeof event.from === "string" ? event.from : void 0
+      });
       return;
     }
   }
@@ -625,6 +690,15 @@ var Dispatcher = class {
   /** Acquire a concurrency slot, run a worker, release the slot. */
   async spawnWorker(account, prompt, ctx) {
     await this.acquireSlot();
+    const workerId = `${account.id}:${ctx.kind}:${ctx.uid ?? ctx.taskId ?? ""}:${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+    let workerResult = null;
+    this.postActivity("/dispatcher/worker-started", {
+      workerId,
+      agentName: account.name,
+      agentEmail: account.email,
+      kind: ctx.kind,
+      trigger: { uid: ctx.uid, taskId: ctx.taskId, subject: ctx.subject, from: ctx.from }
+    });
     try {
       const { body } = loadPersonaForAgent({
         agent: account,
@@ -634,7 +708,7 @@ var Dispatcher = class {
       });
       this.log("info", `[dispatcher] waking "${account.name}" \u2014 ${ctx.kind}${ctx.taskId ? " " + ctx.taskId : ctx.uid ? " uid=" + ctx.uid : ""}`);
       const mcpEnv = await this.buildMcpEnv();
-      await runWorker(
+      workerResult = await runWorker(
         this.query,
         body,
         prompt,
@@ -647,6 +721,40 @@ var Dispatcher = class {
       );
     } finally {
       this.releaseSlot();
+      const ok = workerResult?.ok === true;
+      const preview = workerResult?.ok ? workerResult.text : workerResult ? workerResult.error : "worker did not start";
+      this.postActivity("/dispatcher/worker-finished", {
+        workerId,
+        agentName: account.name,
+        ok,
+        resultPreview: typeof preview === "string" ? preview.slice(0, 240) : void 0
+      });
+    }
+  }
+  /**
+   * Fire-and-forget POST to the API's worker-activity endpoints.
+   *
+   * Failures are swallowed deliberately — the dispatcher must never
+   * block worker spawn or interrupt teardown because the API is briefly
+   * unreachable. The activity registry is best-effort observability, not
+   * load-bearing state.
+   */
+  postActivity(path, body) {
+    const url = `${this.cfg.apiUrl.replace(/\/$/, "")}/api/agenticmail${path}`;
+    try {
+      const result = this.fetchImpl(url, {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          "Authorization": `Bearer ${this.cfg.masterKey}`
+        },
+        body: JSON.stringify(body)
+      });
+      if (result && typeof result.catch === "function") {
+        void result.catch(() => {
+        });
+      }
+    } catch {
     }
   }
   /** Build the env block we pass to the worker's MCP server child process. */

package/dist/cli.js

@@ -1,17 +1,18 @@
 #!/usr/bin/env node
+import {
+  uninstall
+} from "./chunk-6SCQVJBL.js";
 import {
   install
-} from "./chunk-52LXPWO7.js";
+} from "./chunk-DI3DIMUQ.js";
+import "./chunk-3ZIHOOA4.js";
 import {
   status
-} from "./chunk-V3QMDNTR.js";
-import {
-  uninstall
-} from "./chunk-CQLUFM7N.js";
+} from "./chunk-O4H76K3B.js";
 import "./chunk-US5FT2UB.js";
 import {
   AgenticMailApiError
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 
 // src/cli.ts
 var GREEN = (s) => `\x1B[32m${s}\x1B[0m`;

package/dist/{config-BegnlyPD.d.ts → config-CjEDSvVy.d.ts}

@@ -21,6 +21,13 @@ interface ClaudeCodeIntegrationConfig {
     masterKey: string;
     /** Path to Claude Code's user-level config (typically ~/.claude.json). */
     claudeConfigPath: string;
+    /**
+     * Path to Claude Code's user-level settings (typically
+     * ~/.claude/settings.json) where hooks are registered. Different
+     * file from `claudeConfigPath` — Claude Code splits OAuth/MCP state
+     * from preference/hook state, and so do we.
+     */
+    claudeSettingsPath: string;
     /** Directory where per-agent Claude Code subagent .md files live (typically ~/.claude/agents). */
     agentsDir: string;
     /** Key under mcpServers in Claude Code's config. */
@@ -105,6 +112,7 @@ interface ResolveConfigOptions {
     apiUrl?: string;
     masterKey?: string;
     claudeConfigPath?: string;
+    claudeSettingsPath?: string;
     agentsDir?: string;
     mcpServerName?: string;
     bridgeAgentName?: string;

package/dist/dispatcher-bin.js

@@ -1,8 +1,8 @@
 #!/usr/bin/env node
 import {
   Dispatcher
-} from "./chunk-FBO6F4IC.js";
-import "./chunk-3D5VXS5Y.js";
+} from "./chunk-YWUZKKQ5.js";
+import "./chunk-SBP7MJP2.js";
 
 // src/dispatcher-bin.ts
 async function main() {

package/dist/dispatcher.d.ts

@@ -1,4 +1,4 @@
-import { R as ResolveConfigOptions, A as AgenticMailAccount } from './config-BegnlyPD.js';
+import { R as ResolveConfigOptions, A as AgenticMailAccount } from './config-CjEDSvVy.js';
 
 /**
  * AgenticMail → Claude Code event dispatcher.
@@ -62,6 +62,12 @@ interface SSEEvent {
     taskType?: string;
     task?: string;
     assignee?: string;
+    /**
+     * Optional wake allowlist set by the sender via `send_email({ wake })`.
+     * When present, only listed agents (case-insensitive bare name) get a
+     * Claude turn. When absent, every CC'd recipient wakes (v0.8.x default).
+     */
+    wakeAllowlist?: string[];
     [key: string]: unknown;
 }
 interface DispatcherOptions extends ResolveConfigOptions {
@@ -203,6 +209,15 @@ declare class Dispatcher {
     private streamOne;
     /** Acquire a concurrency slot, run a worker, release the slot. */
     private spawnWorker;
+    /**
+     * Fire-and-forget POST to the API's worker-activity endpoints.
+     *
+     * Failures are swallowed deliberately — the dispatcher must never
+     * block worker spawn or interrupt teardown because the API is briefly
+     * unreachable. The activity registry is best-effort observability, not
+     * load-bearing state.
+     */
+    private postActivity;
     /** Build the env block we pass to the worker's MCP server child process. */
     private buildMcpEnv;
     private acquireSlot;

package/dist/dispatcher.js

@@ -1,7 +1,7 @@
 import {
   Dispatcher
-} from "./chunk-FBO6F4IC.js";
-import "./chunk-3D5VXS5Y.js";
+} from "./chunk-YWUZKKQ5.js";
+import "./chunk-SBP7MJP2.js";
 export {
   Dispatcher
 };

package/dist/http-routes.js

@@ -1,11 +1,12 @@
 import {
   createIntegrationRoutes
-} from "./chunk-WAUWKOHA.js";
-import "./chunk-52LXPWO7.js";
-import "./chunk-V3QMDNTR.js";
-import "./chunk-CQLUFM7N.js";
+} from "./chunk-5H46YKCX.js";
+import "./chunk-6SCQVJBL.js";
+import "./chunk-DI3DIMUQ.js";
+import "./chunk-3ZIHOOA4.js";
+import "./chunk-O4H76K3B.js";
 import "./chunk-US5FT2UB.js";
-import "./chunk-3D5VXS5Y.js";
+import "./chunk-SBP7MJP2.js";
 export {
   createIntegrationRoutes
 };

package/dist/index.d.ts

@@ -1,8 +1,8 @@
 export { install } from './install.js';
 export { UninstallOptions, uninstall } from './uninstall.js';
 export { status } from './status.js';
-import { A as AgenticMailAccount } from './config-BegnlyPD.js';
-export { C as ClaudeCodeIntegrationConfig, I as InstallResult, a as InstallStatus, R as ResolveConfigOptions, U as UninstallResult, r as resolveConfig } from './config-BegnlyPD.js';
+import { A as AgenticMailAccount } from './config-CjEDSvVy.js';
+export { C as ClaudeCodeIntegrationConfig, I as InstallResult, a as InstallStatus, R as ResolveConfigOptions, U as UninstallResult, r as resolveConfig } from './config-CjEDSvVy.js';
 export { createIntegrationRoutes } from './http-routes.js';
 export { Dispatcher, DispatcherOptions, QueryFn } from './dispatcher.js';
 import 'express';

package/dist/index.js

@@ -1,19 +1,20 @@
 import {
   Dispatcher,
   loadPersonaForAgent
-} from "./chunk-FBO6F4IC.js";
+} from "./chunk-YWUZKKQ5.js";
 import {
   createIntegrationRoutes
-} from "./chunk-WAUWKOHA.js";
+} from "./chunk-5H46YKCX.js";
+import {
+  uninstall
+} from "./chunk-6SCQVJBL.js";
 import {
   install
-} from "./chunk-52LXPWO7.js";
+} from "./chunk-DI3DIMUQ.js";
+import "./chunk-3ZIHOOA4.js";
 import {
   status
-} from "./chunk-V3QMDNTR.js";
-import {
-  uninstall
-} from "./chunk-CQLUFM7N.js";
+} from "./chunk-O4H76K3B.js";
 import "./chunk-US5FT2UB.js";
 import {
   AgenticMailApiError,
@@ -26,7 +27,7 @@ import {
   renderPersonaBody,
   renderSubagentMarkdown,
   resolveConfig
-} from "./chunk-3D5VXS5Y.js";
+} from "./chunk-SBP7MJP2.js";
 export {
   AgenticMailApiError,
   Dispatcher,

package/dist/install.d.ts

@@ -1,4 +1,4 @@
-import { R as ResolveConfigOptions, I as InstallResult, A as AgenticMailAccount, C as ClaudeCodeIntegrationConfig } from './config-BegnlyPD.js';
+import { R as ResolveConfigOptions, I as InstallResult, A as AgenticMailAccount, C as ClaudeCodeIntegrationConfig } from './config-CjEDSvVy.js';
 
 /**
  * Install AgenticMail into Claude Code.

package/dist/install.js

@@ -1,9 +1,10 @@
 import {
   install,
   selectExposableAgents
-} from "./chunk-52LXPWO7.js";
+} from "./chunk-DI3DIMUQ.js";
+import "./chunk-3ZIHOOA4.js";
 import "./chunk-US5FT2UB.js";
-import "./chunk-3D5VXS5Y.js";
+import "./chunk-SBP7MJP2.js";
 export {
   install,
   selectExposableAgents

package/dist/mail-hook.d.ts

@@ -0,0 +1 @@
+#!/usr/bin/env node

package/dist/mail-hook.js

@@ -0,0 +1,147 @@
+#!/usr/bin/env node
+
+// src/mail-hook.ts
+import { readFileSync, writeFileSync, existsSync, mkdirSync } from "fs";
+import { homedir } from "os";
+import { join, dirname } from "path";
+var AGENTICMAIL_DIR = join(homedir(), ".agenticmail");
+var CONFIG_PATH = join(AGENTICMAIL_DIR, "config.json");
+var CURSOR_PATH = join(AGENTICMAIL_DIR, "claudecode-hook-cursor.json");
+var HOOK_VERSION = "1";
+var HTTP_TIMEOUT_MS = 2e3;
+var PRE_TOOL_USE_THROTTLE_MS = 3e4;
+async function readStdinJson() {
+  if (process.stdin.isTTY) return null;
+  return new Promise((resolve) => {
+    let buf = "";
+    process.stdin.setEncoding("utf-8");
+    process.stdin.on("data", (chunk) => {
+      buf += chunk;
+    });
+    process.stdin.on("end", () => {
+      if (!buf.trim()) {
+        resolve(null);
+        return;
+      }
+      try {
+        resolve(JSON.parse(buf));
+      } catch {
+        resolve(null);
+      }
+    });
+    process.stdin.on("error", () => resolve(null));
+    setTimeout(() => resolve(null), 200).unref();
+  });
+}
+async function main() {
+  const input = await readStdinJson();
+  const eventName = input?.hook_event_name ?? "UserPromptSubmit";
+  if (!existsSync(CONFIG_PATH)) return;
+  let cfg;
+  try {
+    cfg = JSON.parse(readFileSync(CONFIG_PATH, "utf-8"));
+  } catch {
+    return;
+  }
+  if (!cfg.masterKey) return;
+  const apiHost = cfg.api?.host ?? "127.0.0.1";
+  const apiPort = cfg.api?.port ?? 3829;
+  const apiUrl = `http://${apiHost}:${apiPort}`;
+  let bridge;
+  try {
+    const r = await fetch(`${apiUrl}/api/agenticmail/accounts`, {
+      headers: { Authorization: `Bearer ${cfg.masterKey}` },
+      signal: AbortSignal.timeout(HTTP_TIMEOUT_MS)
+    });
+    if (!r.ok) return;
+    const data = await r.json();
+    bridge = (data.agents ?? []).find(
+      (a) => a.name === "claudecode" || a.name === "claude" || a.role === "bridge"
+    );
+  } catch {
+    return;
+  }
+  if (!bridge?.apiKey) return;
+  let cursorMs = 0;
+  let lastCheckedMs = 0;
+  if (existsSync(CURSOR_PATH)) {
+    try {
+      const c = JSON.parse(readFileSync(CURSOR_PATH, "utf-8"));
+      if (typeof c?.lastSeenMs === "number") cursorMs = c.lastSeenMs;
+      if (typeof c?.lastCheckedMs === "number") lastCheckedMs = c.lastCheckedMs;
+    } catch {
+    }
+  }
+  const now = Date.now();
+  if (eventName === "PreToolUse" && now - lastCheckedMs < PRE_TOOL_USE_THROTTLE_MS) {
+    return;
+  }
+  let messages = [];
+  try {
+    const r = await fetch(`${apiUrl}/api/agenticmail/mail/inbox?limit=20`, {
+      headers: { Authorization: `Bearer ${bridge.apiKey}` },
+      signal: AbortSignal.timeout(HTTP_TIMEOUT_MS)
+    });
+    if (!r.ok) return;
+    const data = await r.json();
+    messages = data.messages ?? [];
+  } catch {
+    return;
+  }
+  const newOnes = messages.filter((m) => {
+    if (!m.date) return false;
+    const t = new Date(m.date).getTime();
+    return Number.isFinite(t) && t > cursorMs;
+  });
+  if (newOnes.length === 0) {
+    if (eventName === "PreToolUse") {
+      try {
+        if (!existsSync(dirname(CURSOR_PATH))) mkdirSync(dirname(CURSOR_PATH), { recursive: true });
+        writeFileSync(
+          CURSOR_PATH,
+          JSON.stringify({ lastSeenMs: cursorMs, lastCheckedMs: now, hookVersion: HOOK_VERSION }, null, 2)
+        );
+      } catch {
+      }
+    }
+    return;
+  }
+  newOnes.sort((a, b) => new Date(b.date).getTime() - new Date(a.date).getTime());
+  const lines = [];
+  lines.push(`[AgenticMail bridge inbox] You have ${newOnes.length} new email${newOnes.length === 1 ? "" : "s"} since your last turn:`);
+  for (const m of newOnes) {
+    const fromAddr = m.from?.[0]?.address ?? "unknown";
+    const fromName = m.from?.[0]?.name ?? "";
+    const fromDisp = fromName && fromName !== fromAddr ? `${fromName} <${fromAddr}>` : fromAddr;
+    const subj = m.subject ?? "(no subject)";
+    const preview = (m.preview ?? "").replace(/\s+/g, " ").trim().slice(0, 120);
+    const tail = preview ? ` \u2014 ${preview}${preview.length === 120 ? "\u2026" : ""}` : "";
+    lines.push(`- UID ${m.uid} | from ${fromDisp} | "${subj}"${tail}`);
+  }
+  lines.push("");
+  lines.push(
+    "These are real replies from sub-agents (or mid-task questions). If any of them are addressed to you (Claude Code, the host), surface them to the user and act on whichever they direct you to. Use mcp__agenticmail__read_email for the full body, mcp__agenticmail__reply_email (with replyAll: true) to respond on the thread. You do NOT need to ping the user \u2014 just be aware these landed."
+  );
+  const newestMs = Math.max(...newOnes.map((m) => new Date(m.date).getTime()));
+  try {
+    if (!existsSync(dirname(CURSOR_PATH))) mkdirSync(dirname(CURSOR_PATH), { recursive: true });
+    writeFileSync(
+      CURSOR_PATH,
+      JSON.stringify(
+        { lastSeenMs: newestMs, lastCheckedMs: now, hookVersion: HOOK_VERSION },
+        null,
+        2
+      )
+    );
+  } catch {
+  }
+  process.stdout.write(JSON.stringify({
+    hookSpecificOutput: {
+      hookEventName: eventName,
+      additionalContext: lines.join("\n")
+    }
+  }));
+}
+main().catch(() => {
+  process.exit(0);
+});

package/dist/status.d.ts

@@ -1,4 +1,4 @@
-import { R as ResolveConfigOptions, a as InstallStatus } from './config-BegnlyPD.js';
+import { R as ResolveConfigOptions, a as InstallStatus } from './config-CjEDSvVy.js';
 
 /**
  * Inspect the current install state of @agenticmail/claudecode.

package/dist/status.js

@@ -1,8 +1,8 @@
 import {
   status
-} from "./chunk-V3QMDNTR.js";
+} from "./chunk-O4H76K3B.js";
 import "./chunk-US5FT2UB.js";
-import "./chunk-3D5VXS5Y.js";
+import "./chunk-SBP7MJP2.js";
 export {
   status
 };

package/dist/uninstall.d.ts

@@ -1,4 +1,4 @@
-import { R as ResolveConfigOptions, U as UninstallResult } from './config-BegnlyPD.js';
+import { R as ResolveConfigOptions, U as UninstallResult } from './config-CjEDSvVy.js';
 
 /**
  * Uninstall AgenticMail from Claude Code.

package/dist/uninstall.js

@@ -1,8 +1,9 @@
 import {
   uninstall
-} from "./chunk-CQLUFM7N.js";
+} from "./chunk-6SCQVJBL.js";
+import "./chunk-3ZIHOOA4.js";
 import "./chunk-US5FT2UB.js";
-import "./chunk-3D5VXS5Y.js";
+import "./chunk-SBP7MJP2.js";
 export {
   uninstall
 };

package/package.json

@@ -1,13 +1,14 @@
 {
   "name": "@agenticmail/claudecode",
-  "version": "0.1.9",
+  "version": "0.1.12",
   "description": "Claude Code integration for AgenticMail — surfaces every AgenticMail agent as a native Claude Code subagent so any Claude Code session can delegate to them with the Agent tool",
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
   "bin": {
     "agenticmail-claudecode": "dist/cli.js",
-    "agenticmail-claudecode-dispatcher": "dist/dispatcher-bin.js"
+    "agenticmail-claudecode-dispatcher": "dist/dispatcher-bin.js",
+    "agenticmail-mail-hook": "dist/mail-hook.js"
   },
   "exports": {
     ".": {
@@ -40,7 +41,7 @@
     "LICENSE"
   ],
   "scripts": {
-    "build": "tsup src/index.ts src/cli.ts src/install.ts src/uninstall.ts src/status.ts src/http-routes.ts src/dispatcher.ts src/dispatcher-bin.ts --format esm --dts --clean",
+    "build": "tsup src/index.ts src/cli.ts src/install.ts src/uninstall.ts src/status.ts src/http-routes.ts src/dispatcher.ts src/dispatcher-bin.ts src/mail-hook.ts --format esm --dts --clean",
     "test": "vitest run --passWithNoTests",
     "preuninstall": "node scripts/uninstall.mjs",
     "prepublishOnly": "npm run build"