ai-agentmanager 1.4.0 → 1.6.0

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## 1.6.0 (2026-10-31)
+
+- **Changes explained one by one:** `agentmgr doc <agent> walkthrough` explains each hunk of an agent's changes (what, why, risk, what to check) in the order a reviewer should read them, from the same diff the review screen shows. The review screen and a split view for it come next.
+- **Start at login:** a switch in ☰ → Settings installs (or removes) the login service, so the daemon and running agents come back by themselves after a restart, without a terminal command. Getting started lists it until it's on, and `agentmgr ui` mentions it when it's off. Same service as `agentmgr service install` (launchd on macOS, systemd on Linux).
+- **Attachments:** **📎 Attach files** in New agents and Ask Agent: pick, drop or paste screenshots, logs and documents. Each agent gets its own copy, listed in its request, in a folder Claude Code and Codex can read without a prompt (a team or race shares one copy). Backlog items keep theirs until they start. CLI: `--file-attach <path>` on `agentmgr add` and `agentmgr ask`.
+- **Ask Agent:** **Ask Claude** is now **Ask Agent**, with an Engine choice (Claude Code, Codex, Antigravity, Muse Code) and the engine's notice, if any. `agentmgr ask --engine <engine>`.
+
+## 1.5.0 (2026-10-31)
+
+- **Muse Code engine:** Meta's Muse Code CLI (`muse`) as a fourth engine: `--engine muse` or **Muse Code** in the Engine picker. Session adoption, resume (messages typed once idle), conversation and usage, trust and approval questions in the inbox, a clean stop, and Explain. The picker warns when Muse is in contributor mode. Stuck detection from commands and edits isn't there yet for Muse.
+- **Terminals:** reloading the page no longer types the terminal's answers to old color and cursor queries into the program (seen with Muse Code), in agents' terminals and the Terminal. The agent's terminal tab is named after its engine (Codex, Antigravity, Muse Code).
+- **One agent or one per line:** New agents asks whether each line starts its own agent (a list of tickets) or the whole text goes to one agent (a long description, kept as written). The button shows how many agents it will create, and the choice is remembered.
+- **Diagrams:** a **Diagrams** button beside the documents draws the problem and its solution for a quick grasp, diagrams only: the problem, architecture (changes highlighted), design, data flow with animated arrows, the key interaction and the plan. Drawn from the analysis, design, flow, proposal and plan already written, plus the agent's work and the repo layout. Also `agentmgr doc <agent> diagrams`.
+- **Home page:** shows what you can do (agents, teams and races, Inbox, Code, shells and the browser, Terminal, review and Land, Backlog and the assistant, Results, your phone, several machines), each with a button to it, and stays after Getting started is done or hidden. It scrolls when it doesn't fit.
+- **Restart from the app's own terminal:** `agentmgr restart` or `agentmgr update` typed in an agent's shell or the Terminal no longer dies halfway (daemon stopped, never started) when the restart closes that terminal; it finishes, and the terminal can be reopened.
+- **Remote typing:** over remote access, typing or a paste whose request fails (a network blip) is retried in order instead of dropped, and you're told if it still can't get through.
+
 ## 1.4.0 (2026-10-31)
 
 - **Move an agent to another machine:** **Move to…** on the hub (or `agentmgr fleet move <machine>/<agent> --to <machine>`) continues an agent elsewhere with its conversation, branch and uncommitted work. The work goes through the repo's remote on a temporary handoff branch; the conversation goes through the hub in pieces (not stored). Both machines must allow the new `move` permission. A failed move puts everything back; the copy left behind is marked moved and can't be resumed.

package/README.md

@@ -6,7 +6,7 @@ Run one background [Claude Code](https://claude.com/claude-code) agent per inten
 
 ![The conflict radar: two agents changed the same lines of README.md; the AI judge rates it a collision and suggests which lands first](docs/images/conflict-radar.jpg)
 
-- **Intents in, agents out.** Give one intent or a list (one per line). Each becomes its own `claude` process.
+- **Intents in, agents out.** Give a list, one per line (say, a ticket each), and each becomes its own `claude` process. Or pick **One agent for all of it** and a long description goes to a single agent as written.
 - **Repo or folder.** Give a git URL (cloned once, then each agent gets its own worktree and branch) or a local folder.
   - **Fast clones.** URLs are cloned as partial clones (`--filter=blob:none`): history now, file contents only for what's checked out.
   - **Progress.** Clone and checkout progress shows in the UI and `agentmgr ls`.
@@ -42,7 +42,7 @@ Run one background [Claude Code](https://claude.com/claude-code) agent per inten
 - **Follows your conventions.** Before changing anything, agents review `CLAUDE.md` / `AGENTS.md`, skills, hooks and settings (repo and `~/.claude`), and follow them.
 - **Tokens and cost.** Input, output and cache tokens are tracked per session, subagents included, alongside the cost Claude Code reports.
 - **No trust prompt.** Folders agents run in are pre-trusted, so Claude's "trust this folder" prompt doesn't block them.
-- **Survives restarts and reboots.** Agents that were running when the daemon stopped (restart, crash or reboot) are resumed with the same session IDs when it starts again. `agentmgr service install` starts the daemon at login.
+- **Survives restarts and reboots.** Agents that were running when the daemon stopped (restart, crash or reboot) are resumed with the same session IDs when it starts again. For that after a reboot, turn on **Start at login** in ☰ → Settings (or run `agentmgr service install`); without it, the daemon starts only when you run `agentmgr`.
 - **Shells next to each agent.** **Open shell** gives a terminal in the agent's folder for running commands by hand. **+ Shell** opens as many more as you need, each in its own tab.
 - **Browser tabs.** **+ Browser** opens a page as this machine sees it, in a tab next to the agent's terminals. That can be the agent's dev server on `localhost:5173`, a site on its network or VPN, or any address. Local addresses the agent prints are offered as suggestions. It works from your phone or laptop too (see below).
 - **Live browser.** Browser tabs run a real Chrome on this machine and stream it into the tab: you see its screen and your clicks, scrolling, typing and paste go back to it, from the laptop or the phone. Because the page runs here, `localhost` is this machine, sites that refuse to be framed load normally, and popups (a sign-in window) open as tabs of their own and come back when they close. Chrome (or Chromium, Brave, Edge; `browserBin` in config.json picks one) runs headless with its own profile in `~/.agentmanager/browser-profile`, so sign-ins are remembered; it starts on the first view and stops after 15 minutes with nobody watching. Downloads go to `~/Downloads/agentmanager`. A site that still refuses it as a bot may accept `"browserHeadless": false`, which runs Chrome with a real window on this machine's screen. On Linux where Chrome's sandbox can't start (containers, Ubuntu 23.10+ user-namespace restrictions) the tab says so; `"browserNoSandbox": true` runs it without the sandbox, at your own risk. **Lite** on the bar switches a tab to the frame mode (the page loads in your own browser; quicker for a dev server).
@@ -51,7 +51,8 @@ Run one background [Claude Code](https://claude.com/claude-code) agent per inten
 - **Your layout.** Drag the edge of the agent list, or the line between the brief and the terminal, to resize them; double-click a line to reset it. **Expand** (Ctrl+Shift+E) fills the window with the terminal, shell or browser tab; **Full screen** fills the screen. Agents active in the last day sort to the top; idle ones are dimmed.
 - **As an app.** In Chrome or Edge, click the install icon in the address bar (or ⋮ → Cast, save and share → Install page as app). AI Agent Manager then opens in its own window with its own Dock or taskbar icon, and the daemon keeps running as before.
 - **About.** The version next to the title opens the About box: release date, author, license, links, and how this daemon runs (data folder, port, Node.js, which `claude`). **Check for updates** asks npm for the newest version. In a terminal: `agentmgr about` (`--check` for updates).
-- **General sessions.** **Ask Claude** starts a session with no repo, folder or intent, for questions. It is tracked and resumable like any other agent.
+- **General sessions.** **Ask Agent** starts a session with no repo, folder or intent, for questions, on Claude Code or any other engine you have (Codex, Antigravity, Muse Code). It is tracked and resumable like any other agent.
+- **Attachments.** **📎 Attach files** in New agents and Ask Agent (or drop files on the dialog, or paste a screenshot into the text box) adds a screenshot, a log or a document to the request. Each agent gets its own copy in `~/.agentmanager/uploads/<agent>/`, listed in its request, and Claude Code and Codex can read that folder without asking. From the CLI: `agentmgr add … --file-attach error.log` and `agentmgr ask … --file-attach shot.png`.
 
 New in 0.2, for running many agents at once:
 
@@ -83,13 +84,13 @@ Runs on macOS and Linux; on Windows, use WSL2. Requires Node 20+, `git`, and the
 
 ```sh
 npm install -g ai-agentmanager          # the latest release
-npm install -g ai-agentmanager@1.4.0    # or a specific one
+npm install -g ai-agentmanager@1.6.0    # or a specific one
 ```
 
 Or from the repository, at a release tag:
 
 ```sh
-git clone --branch v1.4.0 --depth 1 https://github.com/jdmunta/ai-agentmanager-core.git
+git clone --branch v1.6.0 --depth 1 https://github.com/jdmunta/ai-agentmanager-core.git
 cd ai-agentmanager-core
 npm install
 npm link            # optional: puts `agentmgr` on your PATH
@@ -109,7 +110,7 @@ npm link
 | Installed with | Update |
 | --- | --- |
 | npm | `npm install -g ai-agentmanager@latest` |
-| A clone at a release tag | `git fetch --tags && git checkout v1.4.0 && npm install` (the tag you want) |
+| A clone at a release tag | `git fetch --tags && git checkout v1.6.0 && npm install` (the tag you want) |
 | A clone of `main` | `git pull && npm install` |
 
 **Check your setup.** `agentmgr doctor` checks Node.js, Claude Code and its sign-in, git, `gh`, the data folder, the daemon's port and Chrome, and says how to fix anything missing.
@@ -327,6 +328,7 @@ The Brief panel's **Share for review** buttons, or `agentmgr doc <agent> <kind>`
 | **Flow** | Summary, end-to-end flowchart with a step-by-step walk-through, key interactions as a sequence diagram, failure and edge cases |
 | **Proposal** | Summary, problem, recommendation, scope, plan, effort and dependencies, risks, decisions needed |
 | **Human vs AI** | Your top 5 decisions, each with why it needs you, the stakes, options and what the work suggests, plus a box for your answer. Up to 5 questions the AI answers, each with its confidence, evidence and how to check it. A second model's verdict on each. |
+| **Diagrams** | Diagrams only, for a quick grasp: the problem, the architecture (with what changes highlighted), the design, the data flow (animated) and the key interaction, plus the plan. Drawn from the documents already written and the agent's work. Optional: **Write all** leaves it out. |
 
 **Human vs AI.** This review takes two separate read-only calls.
 1. **The split.** The agent's engine splits the open questions. *For you to decide* covers judgment calls such as intent, priorities, risk appetite and trade-offs. *Answered by AI* covers reasoning and evaluation: correctness, edge cases, test coverage and what's left.
@@ -723,12 +725,14 @@ Or use **Remote access** in the header of the UI on this computer.
 - **Firewall.** If the macOS firewall is off, the port is open to your whole network while remote access is on. Turn it off when you don't need it.
 - **Streaming.** The phone UI streams with Server-Sent Events plus plain HTTPS requests. iOS Safari doesn't apply an accepted self-signed certificate to WebSockets.
 
-## Engines: Claude Code, Codex and Antigravity
+## Engines: Claude Code, Codex, Antigravity and Muse Code
 
 Agents run Claude Code by default. With OpenAI's Codex CLI installed (`npm i -g @openai/codex`, signed in), an agent can run Codex instead: `agentmgr add --engine codex …`, or the **Engine** picker in New agent (shown once a second engine is installed). Everything else works the same way: worktrees, the brief and questions, the inbox (Codex's approval menus included), token usage, stuck detection, PR tracking, Explain, the backlog and the hub. Codex records no cost, and planning while a repo clones and forks are Claude-only for now. Set `"engine": "codex"` in config.json to make it the default; `engines.codex.bin` points at another binary. On the hub, each machine reports the engines it has (`engine:codex`), so shared backlog work can ask for one. See [docs/engines.md](docs/engines.md).
 
 Google's **Antigravity CLI** (`agy`, signed in with your Google account) works the same way: `--engine antigravity`. It replaced Gemini CLI for personal Google accounts in June 2026. AI Agent Manager adopts the conversation agy starts, resumes it with `--conversation`, reads token usage, commands and file writes from its transcript, shows its permission menus (commands, file changes, access outside the folder) in the inbox, pre-trusts agent folders in agy's settings, and runs Explain in agy's print mode. agy asks before writing the brief too, unless the agent runs in `acceptEdits` mode. `engines.antigravity.bin` points at another binary.
 
+Meta's **Muse Code** (`muse`, installed with `curl -fsSL https://dev.meta.ai/install.sh | bash` and signed in with `muse login`) works the same way: `--engine muse`. AI Agent Manager adopts the session Muse starts, resumes it with `muse resume` (a message for a resumed session is typed once Muse is idle), trusts the agent's folder for each run with `--trust-workspace` instead of saving it, reads the conversation and token usage from Muse's session log, shows its trust and approval questions in the inbox, stops it cleanly with SIGTERM, and runs Explain with `muse exec` in the read-only profile. Commands and edits aren't read from its log yet, so stuck detection has less to go on for Muse agents. When Muse runs in contributor mode (an account without a plan), the engine picker says so: Meta may use your code and conversations to improve its products. `engines.muse.bin` points at another binary, and `engines.muse.args` adds arguments to every run.
+
 ## Several machines: the hub
 
 With agents on more than one Mac, a **hub** shows all of them in one place: every machine's agents and one inbox of what needs you. Where a machine allows it, you can also act from the hub: answer, message, start and stop agents, start new ones, and open live terminals. Each Mac keeps running its own daemon. It connects out to the hub, so no ports need opening on it, and the hub's requests travel back over that same connection.
@@ -917,7 +921,7 @@ Usage comes from the session transcripts and is refreshed every few seconds whil
   - **From another device** (remote access), AI Agent Manager gives each site its own HTTPS port. That port relays to the site as the machine sees it, so you can check from your phone what the agent built or what the machine can reach. Absolute links, redirects and hot-reload WebSockets on the same site keep working, and the site's "don't show me in a frame" headers are dropped. A link or redirect to a different site leaves the relay.
   - **Safety:** the relay serves only signed-in devices. It never relays to AI Agent Manager's own ports on any of the machine's addresses, and it removes AI Agent Manager's sign-in cookie before forwarding. Sites other than localhost must present a valid certificate. The relays close when remote access is turned off.
   - If a tab stays blank on a device, open it once with **↗** and accept the certificate there.
-- **Ask Claude** (or `agentmgr ask`) creates a `general` agent. All general sessions share `~/.agentmanager/general`, so Claude's folder-trust prompt only appears the first time.
+- **Ask Agent** (or `agentmgr ask`, with `--engine` for another engine) creates a `general` agent. All general sessions share `~/.agentmanager/general`, so Claude's folder-trust prompt only appears the first time.
 
 ## Configuration
 

package/docs/engines.md

@@ -1,6 +1,6 @@
 # Design: several agent engines (Claude Code, Codex CLI, Gemini CLI)
 
-Status: phase 1, Codex and Antigravity built. Gemini CLI dropped: since June 18, 2026 it no longer serves personal Google accounts, and Antigravity CLI (`agy`, `src/engines/antigravity.js`, checked against 1.4.3, fixtures in `test/fixtures/antigravity/`) replaced it. 0.4 (branch `v0.4-engines`) · 2026-10-18. Targets the current releases of each CLI, checked in phase 1.
+Status: phase 1, Codex, Antigravity and Muse Code built. Gemini CLI dropped: since June 18, 2026 it no longer serves personal Google accounts, and Antigravity CLI (`agy`, `src/engines/antigravity.js`, checked against 1.4.3, fixtures in `test/fixtures/antigravity/`) replaced it. 0.4 (branch `v0.4-engines`) · 2026-10-18. Targets the current releases of each CLI, checked in phase 1.
 
 AI Agent Manager runs every agent with Claude Code. If Anthropic has an outage, a rate limit hits, or prices change, every agent stops at once. Running agents on other engines as well makes the fleet resilient across providers, and lets each task use the engine that suits it.
 
@@ -167,6 +167,7 @@ The e2e suite uses a fake Claude (`test/fake-claude.js`). Each engine gets its o
 | 1. Adapter seam ✅ | `src/engines/claude.js` with today's behaviour behind the interface; `engine` on agents; real CLIs checked and recorded as fixtures |
 | 2. Antigravity CLI ✅ | Replaces Gemini CLI. `src/engines/antigravity.js`: launch with instructions in the first message, adoption of the conversation agy picks (`cache/last_conversations.json`), resume with `--conversation`, usage, commands and writes from `brain/<id>/.system_generated/logs/transcript_full.jsonl`, command / file / access menus, trust in `settings.json` `trustedWorkspaces`, Explain in print mode (`--mode plan`, prompt attached to `--print=`). Non-Claude engines get a shorter conventions note: a smaller model turned the Claude-specific one into a search of the whole disk. |
 | 3. Codex CLI ✅ | `src/engines/codex.js`: launch and resume, session adoption, brief folder (`--add-dir`), usage, stuck detection, prompts, trust, Explain through `codex exec`; engine pickers in the CLI, the page and the hub; machines report `engine:*` labels for the shared backlog. Not yet: planning while cloning and forks. Built before Gemini, which waits on a sign-in. |
+| 3b. Muse Code ✅ (partly) | `src/engines/muse.js`, checked against Muse Code 1.4.14 (fixtures in `test/fixtures/muse/`). Launch with instructions in the first message; adoption of the session Muse picks (`$XDG_DATA_HOME/muse/sessions/YYYY/MM/DD/<id>/session.jsonl`, folder from the `runtime.session.metadata` record's `workspace_root`); `muse resume <id>`, which takes no message, so a message for a resumed session is typed once it is idle (`supports.resumePrompt: false`); `--trust-workspace` on each launch instead of saving trust; Plan → `:read-only` profile, Auto → `:auto-review`, Bypass → `--yolo`; prompts, replies and usage from the run events; stopped with SIGTERM (`stopSignal`), since a hangup leaves no clean end in its log; Explain through `muse exec` in the read-only profile (options after `exec`: `oneShotExtra`). A notice in the engine picker when Muse runs in contributor mode (Meta may use the content). Not yet, because real model calls weren't available while building it: commands and edits from the log (no stuck detection from them), and the approval menus are matched from the wording in the binary rather than a recorded screen. |
 | 4. Fail over | Engine health, preference lists in both backlogs, engines reported to the hub, **Continue on…** handoff |
 
 ## Open questions

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-agentmanager",
-  "version": "1.4.0",
+  "version": "1.6.0",
   "description": "Run background coding agents (Claude Code, Codex) per intent, and keep track of them from a web UI, your phone or the CLI",
   "keywords": [
     "claude",

package/public/app.js

@@ -4,6 +4,7 @@ import { marked } from '/vendor/marked/marked.esm.js';
 import DOMPurify from '/vendor/dompurify/purify.es.mjs';
 import { touchScroll } from '/touchscroll.js';
 import { openStream, post, REMOTE } from '/transport.js';
+import { attachBox } from '/attach.js';
 import { initReview } from '/review.js';
 import { initLand, prChips } from '/land.js';
 import { initTriggers } from '/triggers.js';
@@ -245,7 +246,9 @@ const INSIGHT_ICONS = { why: '🎯', what: '📦', how: '🛠', now: '📍' };
 
 // ------------------------------------------------------------------ stakeholder documents
 
-const DOCS = { analysis: 'Analysis', design: 'Design', flow: 'Flow', proposal: 'Proposal', humanai: 'Human vs AI' };
+const DOCS = { analysis: 'Analysis', design: 'Design', flow: 'Flow', proposal: 'Proposal', humanai: 'Human vs AI', diagrams: 'Diagrams' };
+// Optional extras that "Write all" leaves out.
+const DOC_EXTRAS = new Set(['diagrams']);
 const DOC_TITLES = { ...DOCS, plan: 'Implementation plan' };
 /** What the document dialog shows: a stakeholder document, or the agent's plan under review. */
 const docOf = (a, kind) => (kind === 'plan' ? a?.planReview && { markdown: a.planReview.markdown, at: a.planReview.updatedAt } : a?.docs?.[kind]);
@@ -308,7 +311,7 @@ function docHtml(markdown) {
 }
 
 function docButtons(a) {
-  const missing = Object.keys(DOCS).filter((k) => !a.docs?.[k]?.markdown && !a.docs?.[k]?.pending);
+  const missing = Object.keys(DOCS).filter((k) => !DOC_EXTRAS.has(k) && !a.docs?.[k]?.markdown && !a.docs?.[k]?.pending);
   const all = missing.length
     ? `<button class="small ghost" data-doc-all title="Write ${missing.map((k) => DOCS[k]).join(', ')} in the background; you'll be told when each is ready">Write all in background</button>`
     : '';
@@ -317,7 +320,7 @@ function docButtons(a) {
       const d = a.docs?.[k];
       const stale = d?.markdown && a.brief?.updatedAt && d.briefUpdatedAt !== a.brief.updatedAt;
       return `<button class="small doc-btn ${d?.markdown ? 'has' : ''}" data-doc="${k}" ${d?.pending ? 'disabled' : ''}
-        title="${d?.markdown ? `Open the ${label.toLowerCase()} document${stale ? ' (the brief changed since)' : ''}` : `Write a ${label.toLowerCase()} document for stakeholders`}">${d?.pending ? `${label}…` : label}${d?.markdown ? (stale ? ' •' : ' ✓') : ''}</button>`;
+        title="${k === 'diagrams' ? (d?.markdown ? `Open the diagrams${stale ? ' (the brief changed since)' : ''}` : 'Draw the problem and its solution for a quick grasp: architecture, design, data flow (animated) and the key interaction, diagrams only. Built from the documents already written and the agent\'s work') : d?.markdown ? `Open the ${label.toLowerCase()} document${stale ? ' (the brief changed since)' : ''}` : `Write a ${label.toLowerCase()} document for stakeholders`}">${d?.pending ? `${label}…` : label}${d?.markdown ? (stale ? ' •' : ' ✓') : ''}</button>`;
     })
     .join('')}</span>`;
 }
@@ -332,8 +335,9 @@ function renderDocDialog() {
   const plan = docView.kind === 'plan';
   $('#doc-title').textContent = `${label} · ${a?.name ?? ''}`;
   const body = $('#doc-body');
+  body.classList.toggle('diagrams-view', docView.kind === 'diagrams');
   if (d?.pending) {
-    body.innerHTML = `<p class="hint">${docView.kind === 'humanai' ? 'Splitting the open questions and answering the AI ones, then asking a second model to check…' : `Writing the ${label.toLowerCase()} document from the agent's work so far…`}</p>`;
+    body.innerHTML = `<p class="hint">${docView.kind === 'humanai' ? 'Splitting the open questions and answering the AI ones, then asking a second model to check…' : docView.kind === 'diagrams' ? "Drawing the diagrams from the documents and the agent's work so far…" : `Writing the ${label.toLowerCase()} document from the agent's work so far…`}</p>`;
     body.dataset.rendered = '';
   } else if (d?.error) body.innerHTML = `<p class="error">Couldn't write it: ${esc(d.error)}</p>`;
   else if (plan && !d?.markdown) body.innerHTML = '<p class="hint">The agent is investigating and hasn\'t written its plan yet. You\'ll see it here (and in the inbox) when it has.</p>';
@@ -345,8 +349,8 @@ function renderDocDialog() {
     }
   }
   const stale = !plan && d?.markdown && a.brief?.updatedAt && d.briefUpdatedAt !== a.brief.updatedAt;
-  const by = d?.writer ? ` by ${d.writer}${d.critique ? `, checked by ${d.critique.label}` : ''}` : '';
-  $('#doc-meta').textContent = d?.markdown ? `${plan ? `Plan · ${PLAN_LABEL[a.planReview.status]} · updated` : 'Written'} ${ago(d.at)}${by}${stale ? ' · the brief has changed since; Regenerate for the latest' : ''}` : '';
+  const by = d?.writer ? ` by ${d.writer}${d.critique ? `, checked by ${d.critique.label}` : ''}${d.from ? ` from ${d.from.length ? d.from.join(', ') : "the agent's work"}` : ''}` : '';
+  $('#doc-meta').textContent = d?.markdown ? `${plan ? `Plan · ${PLAN_LABEL[a.planReview.status]} · updated` : docView.kind === 'diagrams' ? 'Drawn' : 'Written'} ${ago(d.at)}${by}${stale ? ' · the brief has changed since; Regenerate for the latest' : ''}` : '';
   for (const id of ['doc-copy', 'doc-download', 'doc-print']) $(`#${id}`).disabled = !d?.markdown;
   $('#doc-print').hidden = plan;
   $('#doc-regen').hidden = plan;
@@ -1784,6 +1788,9 @@ const term = {
     this.xterm.loadAddon(this.fit);
     this.xterm.open($('#terminal'));
     this.xterm.onData((d) => {
+      // Replaying the backlog makes xterm answer the color and cursor queries in it again; those
+      // answers are stale, and the program would get them as typing.
+      if (this.replaying) return;
       lastTyped = Date.now();
       this.send({ t: 'i', d });
     });
@@ -1842,6 +1849,7 @@ const term = {
           this.conn = null;
           if (ev.code === 1008) toast(`Terminal unavailable: ${ev.reason}`, 5000);
         },
+        onInputLost: () => this.conn === conn && toast('Some typing did not reach the computer (connection lost); check the terminal before going on', 8000),
       },
     );
     this.conn = conn;
@@ -1852,7 +1860,8 @@ const term = {
       if (msg.t === 'hello') {
         // Also sent again after a reconnect: start from a clean screen and replay.
         this.xterm.reset();
-        this.xterm.write(msg.backlog);
+        this.replaying = true;
+        this.xterm.write(msg.backlog, () => (this.replaying = false));
         if (!matchMedia('(pointer: coarse)').matches) this.xterm.focus();
       } else if (msg.t === 'error') {
         toast(`Terminal unavailable: ${msg.error}`, 5000);
@@ -1888,6 +1897,8 @@ const term = {
   renderTabs() {
     renderSignIn();
     const a = this.agentId && agents.get(this.agentId);
+    // The agent's own session tab is named after its engine (Codex, Muse Code, …).
+    $('.term-tab[data-kind="claude"]').textContent = ENGINE_LABEL[a?.engine] ?? 'Claude';
     const nums = new Set(a?.runtime.shells ?? []);
     if (isShell(this.kind)) nums.add(shellNum(this.kind));
     $('#shell-tabs').innerHTML = [...nums]
@@ -3137,6 +3148,7 @@ $('#recent-bar').addEventListener('click', (e) => {
 // ------------------------------------------------------------------ new agents dialog
 
 const dialog = $('#new-dialog');
+const newFiles = attachBox({ slot: $('#new-attach'), form: $('#new-form'), paste: $('#new-form').intents, esc, toast });
 function renderProjectOptions() {
   $('#project-options').innerHTML = projectsState
     .map((p) => `<option value="${esc(p.shortcut)}">${esc(p.name)}${p.repo ? ` · ${esc(p.repo)}` : p.dir ? ` · ${esc(p.dir)}` : ''}</option>`)
@@ -3147,6 +3159,7 @@ function openNewDialog(mode = 'now') {
   renderProjectOptions();
   $('#new-error').hidden = true;
   dialog.querySelector(`input[name="when"][value="${mode}"]`).checked = true;
+  restoreIntentSplit();
   syncWhen();
   dialog.showModal();
   dialog.querySelector('textarea').focus();
@@ -3156,7 +3169,55 @@ function syncWhen() {
   dialog.querySelector('.now-only').hidden = when !== 'now';
   dialog.querySelector('.race-only').hidden = when !== 'race';
   dialog.querySelector('.team-only').hidden = when !== 'team';
-  dialog.querySelector('button[type=submit]').textContent = { backlog: 'Add to backlog', race: 'Start race', team: 'Start team' }[when] ?? 'Create agents';
+  // A race or a team always works on the whole text.
+  dialog.querySelector('.intent-split').hidden = when === 'race' || when === 'team';
+  syncIntentPlaceholder();
+  syncIntentCount();
+}
+
+/** The form's intents: one per line, or the whole text as one (kept as written, blank lines and all). */
+function formIntents(form) {
+  const text = form.intents.value;
+  const whole = form.split.value === 'one' || ['race', 'team'].includes(form.when.value);
+  if (whole) return text.trim() ? [text.trim()] : [];
+  return text
+    .split('\n')
+    .map((s) => s.trim())
+    .filter(Boolean);
+}
+
+function syncIntentCount() {
+  const form = $('#new-form');
+  const when = form.when.value;
+  const n = formIntents(form).length;
+  const split = !['race', 'team'].includes(when);
+  $('#intent-count').textContent = split && n ? (when === 'backlog' ? `${n} item${n > 1 ? 's' : ''}` : `${n} agent${n > 1 ? 's' : ''}`) : '';
+  const submit = dialog.querySelector('button[type=submit]');
+  submit.textContent = { backlog: n > 1 ? `Add ${n} to backlog` : 'Add to backlog', race: 'Start race', team: 'Start team' }[when] ?? (n > 1 ? `Create ${n} agents` : 'Create agent');
+}
+$('#new-form').intents.addEventListener('input', syncIntentCount);
+for (const r of dialog.querySelectorAll('input[name="split"]')) {
+  r.addEventListener('change', () => {
+    try {
+      localStorage.setItem('agentmgr.intentSplit', r.value);
+    } catch {}
+    syncIntentPlaceholder();
+    syncIntentCount();
+  });
+}
+const LINES_PLACEHOLDER = $('#new-form').intents.placeholder;
+const WHOLE_PLACEHOLDER = 'Describe the work in as much detail as you like: the problem, what done looks like, constraints, links. It all goes to one agent.';
+/** The remembered choice (a form reset forgets it), and a placeholder to match. */
+function restoreIntentSplit() {
+  const form = $('#new-form');
+  try {
+    form.split.value = localStorage.getItem('agentmgr.intentSplit') === 'one' ? 'one' : 'lines';
+  } catch {}
+  syncIntentPlaceholder();
+}
+function syncIntentPlaceholder() {
+  const form = $('#new-form');
+  form.intents.placeholder = form.split.value === 'one' || ['race', 'team'].includes(form.when.value) ? WHOLE_PLACEHOLDER : LINES_PLACEHOLDER;
 }
 for (const r of dialog.querySelectorAll('input[name="when"]')) r.addEventListener('change', syncWhen);
 $('#new-btn').onclick = () => openNewDialog('now');
@@ -3164,10 +3225,13 @@ $('#new-cancel').onclick = () => dialog.close();
 $('#new-form').addEventListener('submit', async (e) => {
   e.preventDefault();
   const f = new FormData(e.target);
-  const intents = String(f.get('intents'))
-    .split('\n')
-    .map((s) => s.trim())
-    .filter(Boolean);
+  const intents = formIntents(e.target);
+  if (newFiles.pending()) {
+    $('#new-error').textContent = 'Still uploading the attached files; try again in a moment';
+    $('#new-error').hidden = false;
+    return;
+  }
+  const attachments = newFiles.paths().length ? newFiles.paths() : undefined;
   if (f.get('when') === 'team') {
     // "analyst=opus reviewer=opus"
     const models = Object.fromEntries(
@@ -3178,6 +3242,7 @@ $('#new-form').addEventListener('submit', async (e) => {
     );
     try {
       const team = await api('POST', '/teams', {
+        attachments,
         intent: intents.join('\n'),
         critics: Number(f.get('critics')),
         mode: f.get('teamMode') || undefined,
@@ -3197,6 +3262,7 @@ $('#new-form').addEventListener('submit', async (e) => {
       teams.set(team.id, team);
       dialog.close();
       e.target.reset();
+      newFiles.clear();
       syncWhen();
       openTeam(team.id);
       toast(team.mode === 'lead' ? 'Team started: one lead session with sub-agents' : `Team started: ${team.members.length} agents`);
@@ -3218,6 +3284,7 @@ $('#new-form').addEventListener('submit', async (e) => {
       });
     try {
       const race = await api('POST', '/races', {
+        attachments,
         intent: intents.join('\n'),
         variants,
         testCommand: f.get('testCommand') || undefined,
@@ -3232,6 +3299,7 @@ $('#new-form').addEventListener('submit', async (e) => {
       races.set(race.id, race);
       dialog.close();
       e.target.reset();
+      newFiles.clear();
       syncWhen();
       openRace(race.id);
       toast(`Race started: ${race.rows.length} attempts`);
@@ -3244,6 +3312,7 @@ $('#new-form').addEventListener('submit', async (e) => {
   if (f.get('when') === 'backlog') {
     try {
       const added = await api('POST', '/backlog', {
+        attachments,
         items: intents,
         repo: f.get('repo') || undefined,
         dir: f.get('dir') || undefined,
@@ -3259,6 +3328,7 @@ $('#new-form').addEventListener('submit', async (e) => {
       });
       dialog.close();
       e.target.reset();
+      newFiles.clear();
       showTab('backlog');
       toast(`Queued ${added.length} item${added.length > 1 ? 's' : ''}`);
     } catch (err) {
@@ -3269,6 +3339,7 @@ $('#new-form').addEventListener('submit', async (e) => {
   }
   try {
     const created = await api('POST', '/agents', {
+        attachments,
       priority: Number(f.get('priority')),
       intents,
       repo: f.get('repo') || undefined,
@@ -3286,6 +3357,7 @@ $('#new-form').addEventListener('submit', async (e) => {
     created.forEach((a) => agents.set(a.id, a));
     dialog.close();
     e.target.reset();
+      newFiles.clear();
     select(created[0].id);
     toast(`Created ${created.length} agent${created.length > 1 ? 's' : ''}`);
     if (created.length === 1 && created[0].runtime.planning) term.open(created[0].id);
@@ -3362,17 +3434,29 @@ $('#ask-btn').onclick = () => {
   askDialog.querySelector('textarea').focus();
 };
 $('#ask-cancel').onclick = () => askDialog.close();
+const askFiles = attachBox({ slot: $('#ask-attach'), form: $('#ask-form'), paste: $('#ask-form textarea'), esc, toast });
 $('#ask-form').addEventListener('submit', async (e) => {
   e.preventDefault();
   const f = new FormData(e.target);
+  if (askFiles.pending()) {
+    $('#ask-error').textContent = 'Still uploading the attached files; try again in a moment';
+    $('#ask-error').hidden = false;
+    return;
+  }
   try {
     const a = await api('POST', '/general', {
       question: String(f.get('question') || '').trim() || undefined,
       name: String(f.get('name') || '').trim() || undefined,
+      engine: f.get('engine') || undefined,
+      attachments: askFiles.paths().length ? askFiles.paths() : undefined,
     });
     agents.set(a.id, a);
     askDialog.close();
+    // Keep the engine for the next question; clear the rest.
+    const engine = e.target.engine.value;
     e.target.reset();
+    e.target.engine.value = engine;
+    askFiles.clear();
     select(a.id);
     // The shared general folder is ready almost immediately; open the terminal once it runs.
     const t0 = Date.now();
@@ -3951,13 +4035,23 @@ $('#login-form').addEventListener('submit', async (e) => {
 });
 
 // Engines this machine can run; the New agent form offers a choice once there's more than Claude.
-const ENGINE_LABEL = { claude: 'Claude', codex: 'Codex', antigravity: 'Antigravity' };
+const ENGINE_LABEL = { claude: 'Claude', codex: 'Codex', antigravity: 'Antigravity', muse: 'Muse Code' };
 async function loadEngines() {
   try {
     const list = (await api('GET', '/engines')).filter((e) => e.installed && e.id !== 'claude');
-    const sel = document.querySelector('#engine-field select');
-    for (const e of list) sel.append(Object.assign(document.createElement('option'), { value: e.id, textContent: e.label }));
-    $('#engine-field').hidden = !list.length;
+    // Something to know before choosing an engine (Muse Code in contributor mode, say).
+    const notices = Object.fromEntries(list.filter((e) => e.notice).map((e) => [e.id, e.notice]));
+    for (const [field, notice] of [['#engine-field', '#engine-notice'], ['#ask-engine-field', '#ask-engine-notice']]) {
+      const sel = document.querySelector(`${field} select`);
+      for (const e of list) sel.append(Object.assign(document.createElement('option'), { value: e.id, textContent: e.label }));
+      $(field).hidden = !list.length;
+      const show = () => {
+        $(notice).textContent = notices[sel.value] ?? '';
+        $(notice).hidden = !notices[sel.value];
+      };
+      sel.addEventListener('change', show);
+      show();
+    }
   } catch {}
 }
 

package/public/attach.js

@@ -0,0 +1,100 @@
+// Attachments for a request that has no agent yet (New agents, Ask): pick, drop or paste files
+// (a screenshot, a log). Each uploads to the daemon right away and shows as a chip; the form sends
+// the uploaded paths, and the daemon copies them to each agent it creates.
+
+const fmtSize = (n) => (n < 1024 ? `${n} B` : n < 1048576 ? `${(n / 1024).toFixed(0)} KB` : `${(n / 1048576).toFixed(1)} MB`);
+
+/**
+ * `slot` is an element to fill with the button and chips; `form` takes drops; `paste` (a textarea)
+ * takes pasted files. Returns { paths(), pending(), clear() }.
+ */
+export function attachBox({ slot, form, paste, esc, toast }) {
+  const items = []; // { id, name, size, path?, error?, uploading }
+  let seq = 0;
+  slot.classList.add('attach-box');
+  slot.innerHTML = `<button type="button" class="small ghost attach-pick" title="Attach files: a screenshot, a log, a document. You can also drop them here or paste into the text box">📎 Attach files</button>
+    <input type="file" multiple hidden />
+    <span class="attach-chips"></span>`;
+  const input = slot.querySelector('input[type=file]');
+  const chips = slot.querySelector('.attach-chips');
+
+  function render() {
+    chips.innerHTML = items
+      .map(
+        (it) => `<span class="attach-chip ${it.error ? 'fail' : ''}" title="${esc(it.error ?? it.path ?? 'Uploading…')}">${esc(it.name)} <span class="hint">${it.uploading ? 'uploading…' : it.error ? 'failed' : fmtSize(it.size)}</span><button type="button" class="link" data-drop="${it.id}" title="Remove">×</button></span>`,
+      )
+      .join('');
+  }
+
+  async function add(files) {
+    for (const file of files) {
+      const name = file.name || `pasted-${Date.now()}.${(file.type.split('/')[1] || 'bin').replace(/\W/g, '')}`;
+      const it = { id: ++seq, name, size: file.size, uploading: true };
+      items.push(it);
+      render();
+      try {
+        const res = await fetch(`/api/uploads?name=${encodeURIComponent(name)}`, {
+          method: 'POST',
+          credentials: 'same-origin',
+          headers: { 'content-type': 'application/octet-stream', 'x-agentmgr': '1' },
+          body: file,
+        });
+        const data = await res.json().catch(() => ({}));
+        if (!res.ok) throw new Error(data.error || `HTTP ${res.status}`);
+        Object.assign(it, { path: data.path, size: data.size });
+      } catch (err) {
+        it.error = err.message;
+        toast(`Could not attach ${name}: ${err.message}`, 5000);
+      }
+      it.uploading = false;
+      render();
+    }
+  }
+
+  slot.querySelector('.attach-pick').onclick = () => input.click();
+  input.onchange = () => {
+    add([...input.files]);
+    input.value = '';
+  };
+  chips.addEventListener('click', (e) => {
+    const id = Number(e.target.closest('[data-drop]')?.dataset.drop);
+    if (!id) return;
+    items.splice(
+      items.findIndex((it) => it.id === id),
+      1,
+    );
+    render();
+  });
+  form.addEventListener('dragover', (e) => {
+    if ([...(e.dataTransfer?.types ?? [])].includes('Files')) {
+      e.preventDefault();
+      form.classList.add('dropping');
+    }
+  });
+  form.addEventListener('dragleave', (e) => {
+    if (!form.contains(e.relatedTarget)) form.classList.remove('dropping');
+  });
+  form.addEventListener('drop', (e) => {
+    form.classList.remove('dropping');
+    const files = [...(e.dataTransfer?.files ?? [])];
+    if (!files.length) return;
+    e.preventDefault();
+    add(files);
+  });
+  // A pasted screenshot (or copied files) attaches; pasted text stays text.
+  paste?.addEventListener('paste', (e) => {
+    const files = [...(e.clipboardData?.files ?? [])];
+    if (!files.length) return;
+    e.preventDefault();
+    add(files);
+  });
+
+  return {
+    paths: () => items.filter((it) => it.path).map((it) => it.path),
+    pending: () => items.some((it) => it.uploading),
+    clear() {
+      items.length = 0;
+      render();
+    },
+  };
+}

package/public/hub/hub.js

@@ -42,7 +42,7 @@ function toast(msg, bad = false) {
   toast.timer = setTimeout(() => (t.hidden = true), bad ? 6000 : 2500);
 }
 
-const ENGINE_LABEL = { claude: 'Claude Code', codex: 'Codex', antigravity: 'Antigravity' };
+const ENGINE_LABEL = { claude: 'Claude Code', codex: 'Codex', antigravity: 'Antigravity', muse: 'Muse Code' };
 /** Engines a machine reports installed (its engine:* labels). */
 const enginesOf = (runner) => (runner?.labels ?? []).filter((l) => l.startsWith('engine:')).map((l) => l.slice(7));
 /** Fill an engine picker; hidden while Claude Code is the only choice. */