@@ -118,6 +118,7 @@ A deep UX overhaul + correctness pass landed on top of the agentic core, all in
 - **Response action bar** — Retry / Copy / 👍 / 👎 + *model · credits*, with inlined VS Code **codicons** (a self-contained `IC` registry + `codicon()` — no font/CSP changes). All UI emojis replaced by codicons.
 - **Sticky surfaces** — review bar (Keep all / Undo all + `+/-` totals) moved between the log and composer; an **active plan bar** pinned above it (opaque, doesn't scroll behind); a **context-usage meter** in the footer.
 - **Interaction polish** — `ask_user` clickable clarifying questions; shimmer "working" text + animated atom indicator; per-response Copy; stick-to-bottom scroll + jump-to-latest; in-chat 🐛 debug logging; the **account** sign-in/profile panel (M9 client, OAuth-callback wired, BYO-key privacy preserved); neutral context chips with colored type badges; light-theme contrast fixes.
+- **Live terminal for approved commands** — `run_command` now **streams** real stdout/stderr (`cp.spawn`, not buffered `exec`) into a terminal card as it runs, ending with `✓ exit 0 · 1.2s` / `✗ exit N` / `⊘ stopped` / `⚠ timed out`. Each command spawns **`detached`** (own process group) so a **Stop button** (and the main ◼) can `process.kill(-pid)` the whole tree — killing a server *and its children* with no orphans (verified). Backstop timeout is configurable (`atompp.ai.commandTimeout`, `0` = off); the model still receives the capped output as the tool result.
 - **Data-safety fixes** — a truncated `tool_use` (cut off by `max_tokens` mid-arguments) is now finalized with `input:{}` in `providers.js`, so the agent **recovers** instead of 400-ing (`tool_use.input: Field required`); EOL/BOM-tolerant edits; apply-then-review snapshot/Undo hardening; fixed a `fileChip` name collision that rendered the active-file chip as "undefined".
 
 ---
@@ -128,6 +129,7 @@ M0→M4 is where you stop wanting to switch editors. These milestones are what m
 
 **M5 — Close the loop & earn hands-off trust.**
 - **● Diagnostics-driven auto-verify loop** *(M)* — after edits land, auto-run a configurable verify command (typecheck/lint/test) + pull `vscode.languages.getDiagnostics` for touched files, feed failures back into the transcript, self-correct until green (bounded by a round budget). *The capability every 2026 source calls the defining trait of autonomous agents — and we already have `run_command` + diagnostics + a persistent loop, so this makes the agent "finish the job" instead of handing back broken code.*
+- **Background command execution** *(M)* — detect long-running commands (dev servers, watchers, `tail -f`) and run them **detached in the background** instead of blocking the agent loop: the agent gets an immediate "started on :3000 (running in background)" and continues, while the live terminal card keeps streaming and a Stop button stays available; the agent can read recent output on demand. *Today an approved server **blocks** the agent until you Stop it or the backstop timeout fires — the live-terminal + process-group kill (M4+) made stopping reliable, but the agent still can't proceed while a server runs. This is how Cursor/Windsurf handle servers, and it pairs with the diagnostics→fix loop (start server → run tests against it → fix → repeat).*
 - **● Workspace checkpoints / restore timeline** *(M)* — snapshot touched files per agent step, with one-click "Restore to here." *Cursor's checkpoints are explicitly "what makes people comfortable letting the agent make sweeping changes"; `reviewSession.js` already snapshots from disk, so this is mostly promoting it to a per-step stack.*
 - **Prompt transparency + editable system prompts** *(S)* — view and override the real `SYSTEM` prompts and see the exact context sent. *A stated PLAN pillar (§4, from Void), still unbuilt; a privacy/trust wedge against closed tools.*
 - **Real token + cost accounting** *(S)* — ✅ *context-usage half shipped*: `providers.js` now captures real Anthropic `usage` (input/output + cache tokens) from the stream, surfaced as an always-visible footer meter + a warning bar as the window fills (`atompp.ai.contextWindow`). *Remaining:* running $ spend + replace `lmProvider`'s length/4 estimate. *Cost opacity is the #1 criticism of BOTH Cursor and Windsurf — a no-backend editor showing exact per-message cost flips their biggest weakness into ours.*
 
@@ -160,14 +160,49 @@ function isBinaryFile(abs) {
 	try { const buf = fs.readFileSync(abs); const n = Math.min(buf.length, 8000); for (let i = 0; i < n; i++) { if (buf[i] === 0) { return true; } } return false; } catch { return false; }
 }
 
-function runCommand(root, command) {
+// Run a shell command, STREAMING its output: onChunk(text, 'stdout'|'stderr') fires as it runs (live
+// terminal in the chat), onExit(code, ms, how) at the end. onStart(child, stop) exposes a stop() that
+// kills the whole process GROUP (server + children) — needed because a server is a child of the shell.
+// Returns the (capped) full output for the model.
+function runCommand(root, command, onChunk, onExit, onStart, timeoutMs) {
 	return new Promise((resolve) => {
-		cp.exec(command, { cwd: root, timeout: 60000, maxBuffer: 1 << 20 }, (err, stdout, stderr) => {
-			let out = (stdout || '') + (stderr ? '\n[stderr]\n' + stderr : '');
-			if (err && err.killed) { out += '\n[timed out after 60s]'; }
-			else if (err) { out += '\n[exit ' + (err.code != null ? err.code : '?') + ']'; }
-			resolve(out.slice(0, 8000) || '(no output)');
-		});
+		let out = '', streamed = 0, settled = false, timedOut = false, stopped = false;
+		const STREAM_CAP = 100000;            // max we push to the live terminal view
+		const start = Date.now();
+		const cap = (s, stream) => {
+			out += s;
+			if (out.length > 262144) { out = out.slice(-262144); }   // bound the model buffer's memory
+			if (onChunk && streamed < STREAM_CAP) {
+				const room = STREAM_CAP - streamed;
+				const piece = s.length > room ? s.slice(0, room) : s;
+				onChunk(piece, stream);
+				streamed += piece.length;
+				if (streamed >= STREAM_CAP) { onChunk('\n[output truncated in view]', 'stderr'); }
+			}
+		};
+		let child;
+		// detached:true → the shell becomes its own process-group leader, so process.kill(-pid) kills the
+		// whole tree (the shell AND the server/children it spawned), not just the shell (which orphans the server).
+		try { child = cp.spawn(command, { cwd: root, shell: true, detached: true }); }
+		catch (e) { if (onExit) { onExit(-1, 0, 'error'); } resolve('ERROR: ' + ((e && e.message) || e)); return; }
+		const killGroup = (sig) => { try { if (child.pid) { process.kill(-child.pid, sig); } else { child.kill(sig); } } catch (e) { try { child.kill(sig); } catch (e2) { /* already gone */ } } };
+		const stop = () => { if (settled || stopped) { return; } stopped = true; killGroup('SIGTERM'); setTimeout(() => { if (!settled) { killGroup('SIGKILL'); } }, 1500); };
+		if (onStart) { onStart(child, stop); }
+		if (child.stdout) { child.stdout.on('data', (d) => cap(d.toString(), 'stdout')); }
+		if (child.stderr) { child.stderr.on('data', (d) => cap(d.toString(), 'stderr')); }
+		const to = (timeoutMs && timeoutMs > 0) ? setTimeout(() => { timedOut = true; killGroup('SIGKILL'); }, timeoutMs) : null;
+		const finish = (code) => {
+			if (settled) { return; } settled = true;
+			if (to) { clearTimeout(to); }
+			const ms = Date.now() - start;
+			const how = stopped ? 'stopped' : timedOut ? 'timeout' : 'exit';
+			const exit = stopped ? 130 : (timedOut ? 124 : (code == null ? -1 : code));
+			if (onExit) { onExit(exit, ms, how); }
+			const note = stopped ? '\n[stopped by user]' : timedOut ? '\n[timed out]' : (exit !== 0 ? '\n[exit ' + exit + ']' : '');
+			resolve((out + note).slice(0, 8000) || '(no output)');
+		};
+		child.on('close', (code) => finish(code));
+		child.on('error', (e) => { cap('\n' + ((e && e.message) || e), 'stderr'); finish(1); });
 	});
 }
 
@@ -231,10 +266,17 @@ async function runTool(tu, ctx) {
 			return 'Applied edit to ' + input.path + ' (pending the user\'s Keep/Undo review).';
 		}
 		if (tu.name === 'run_command') {
-			const approved = await ctx.approve({ kind: 'command', command: String(input.command || ''), explanation: input.explanation || '' });
+			const cmd = String(input.command || '');
+			const approved = await ctx.approve({ kind: 'command', command: cmd, explanation: input.explanation || '' });
 			if (!approved) { return 'User skipped this command. Do not retry it.'; }
-			ctx.post({ type: 'agentTool', icon: 'terminal', text: 'ran: ' + input.command });
-			return await runCommand(root, String(input.command || ''));
+			const runId = tu.id || ('run-' + Date.now());
+			ctx.post({ type: 'termRun', id: runId, command: cmd, cwd: path.basename(root) || 'workspace' });
+			const stops = ctx.commandStops;   // shared registry so the Stop button / ■ can kill the process group
+			return await runCommand(root, cmd,
+				(chunk, stream) => ctx.post({ type: 'termOutput', id: runId, chunk: chunk, stream: stream }),
+				(code, ms, how) => { if (stops) { stops.delete(runId); } ctx.post({ type: 'termExit', id: runId, code: code, ms: ms, how: how }); },
+				(child, stop) => { if (stops) { stops.set(runId, stop); } },
+				ctx.commandTimeout);
 		}
 		if (tu.name === 'ask_user') {
 			const questions = Array.isArray(input.questions) ? input.questions : [];
 
@@ -366,6 +366,9 @@ function postActiveFile() {
 	post({ type: 'activeFile', label: path.basename(ed.document.uri.fsPath) + ' · ' + ed.document.lineCount + ' lines' });
 }
 
+/** Running approved commands: runId → stop() that kills the command's process group. */
+const commandStops = new Map();
+
 /** Pending in-chat approval requests, keyed by id, resolved by the webview. */
 const pendingApprovals = new Map();
 let approvalSeq = 0;
@@ -471,6 +474,8 @@ async function agentFlow(text) {
 			approve: requestApproval,           // run_command only
 			ask: requestQuestions,              // ask_user — clickable clarifying questions
 			contextLimit: cfg.get('contextWindow', 200000), // model context window → drives the usage meter
+			commandStops: commandStops,         // runId → stop() (process-group kill); used by Stop button / ■
+			commandTimeout: cfg.get('commandTimeout', 120000), // backstop before a command is force-killed (0 = off)
 			applyEdit: (req) => review.applyEdit(req), // file edits: apply-then-review
 			signal: abort.signal
 		});
@@ -609,7 +614,8 @@ class ChatViewProvider {
 				case 'ready': sendConfigToWebview(); postActiveFile(); postContextFiles(); post({ type: 'mode', agent: agentMode }); postAccount(); buildFileIndex(); post({ type: 'contextUsage', input: 0, limit: aiConfig().get('contextWindow', 200000) }); if (review) { review.resync(); } break;
 				case 'setMode': agentMode = !!msg.agent; post({ type: 'mode', agent: agentMode }); break;
 				case 'send': await handleSend(msg.text); break;
-				case 'stop': dbg('stop.clicked'); if (abort) { abort.abort(); } clearApprovals(); clearQuestions(); break;
+				case 'stop': dbg('stop.clicked', { running: commandStops.size }); for (const [, stop] of commandStops) { try { stop(); } catch (e) { /* gone */ } } if (abort) { abort.abort(); } clearApprovals(); clearQuestions(); break;
+				case 'stopCommand': { dbg('stopCommand', { id: msg.id }); const s = commandStops.get(msg.id); if (s) { try { s(); } catch (e) { /* gone */ } } break; }
 				case 'approvalResponse': resolveApproval(msg.id, msg.approved); break;
 				case 'questionsResponse': resolveQuestions(msg.id, msg.answers, msg.notes); break;
 				case 'accountSignIn': await accountSignIn(msg.provider, msg.create); break;
 
@@ -176,6 +176,21 @@
   .term .termfoot .skip { background: var(--vscode-button-secondaryBackground, rgba(127,127,127,.25)); color: var(--vscode-button-secondaryForeground, var(--vscode-foreground)); }
   .term .verdict { font-size: 12px; color: var(--muted); }
   .term.ran .verdict { color: var(--vscode-terminal-ansiBrightGreen, var(--vscode-terminal-ansiGreen, #7ee787)); }
+  /* live terminal output (streamed stdout/stderr) — flows under the prompt like a real terminal, no box */
+  .termrun .termout { margin: 4px 0 0; padding: 0; border: none; border-radius: 0; background: none; min-height: 0;
+    max-height: 320px; overflow-y: auto; white-space: pre-wrap; word-break: break-word;
+    font-family: var(--vscode-editor-font-family, ui-monospace, monospace); font-size: 12.5px; line-height: 1.5;
+    color: var(--vscode-terminal-foreground, var(--vscode-foreground)); }
+  .termrun .termout:empty { display: none; margin: 0; }
+  .termrun .termerr { color: var(--vscode-terminal-ansiRed, var(--vscode-errorForeground, #f14c4c)); }
+  .termrun .termstatus { margin-top: 9px; font-size: 11.5px; color: var(--muted); display: flex; align-items: center; gap: 6px; }
+  .termrun .termstatus.ok { color: var(--vscode-terminal-ansiBrightGreen, var(--vscode-terminal-ansiGreen, #7ee787)); }
+  .termrun .termstatus.bad { color: var(--vscode-errorForeground, #f14c4c); }
+  .termrun .termstatus .ci { width: 1em; height: 1em; }
+  .termrun .termstop { margin-left: auto; cursor: pointer; border: 1px solid var(--border); border-radius: 5px; padding: 2px 9px; font-size: 11px; display: inline-flex; align-items: center; gap: 4px; background: var(--vscode-button-secondaryBackground, rgba(127,127,127,.22)); color: var(--vscode-foreground); }
+  .termrun .termstop:hover { background: var(--vscode-errorForeground, #f14c4c); color: #fff; border-color: transparent; }
+  .termrun .termstop[disabled] { opacity: .6; cursor: default; }
+  .termrun .termstop .ci { width: 1em; height: 1em; }
 
   /* ---- ask_user: clickable clarifying questions ---- */
   .questions { border: 1px solid var(--border); border-radius: 10px; margin: 6px 2px; padding: 10px 11px 11px; background: var(--field-bg); }
@@ -577,6 +592,7 @@
   log.addEventListener('keydown', function(e){ if (e.key === 'Enter' || e.key === ' '){ const c = e.target.closest && e.target.closest('.filechip'); if (c){ e.preventDefault(); openChip(c); } } });
   const reviewBar = document.getElementById('reviewBar');
   const editCards = {};   // id -> applied-edit card element
+  const termCards = {};   // id -> live terminal-output card element
   reviewBar.querySelector('.keepall').onclick = () => vscode.postMessage({ type: 'reviewKeepAll' });
   reviewBar.querySelector('.undoall').onclick = () => vscode.postMessage({ type: 'reviewUndoAll' });
   const input = document.getElementById('input');
@@ -938,12 +954,48 @@
       vscode.postMessage({ type: 'approvalResponse', id: m.id, approved });
       card.classList.add(approved ? 'ran' : 'skipped');
       const caret = card.querySelector('.tcaret'); if (caret) { caret.remove(); }
-      card.querySelector('.termfoot').innerHTML = '<span class="verdict">' + (approved ? codicon('check') + ' Running in terminal…' : codicon('circle-slash') + ' Skipped') + '</span>';
+      card.querySelector('.termfoot').innerHTML = '<span class="verdict">' + (approved ? codicon('check') + ' Approved' : codicon('circle-slash') + ' Skipped') + '</span>';
     };
     card.querySelector('.approve').onclick = () => done(true);
     card.querySelector('.skip').onclick = () => done(false);
   }
 
+  // Live terminal: an approved command streams its real stdout/stderr here as it runs.
+  function addTermRun(m){
+    clearEmpty(); clearStatus(); agentBubble = null;
+    const card = document.createElement('div'); card.className = 'approval term termrun';
+    card.innerHTML =
+      '<div class="termbar"><span class="tdots"><i class="r"></i><i class="y"></i><i class="g"></i></span>'
+        + '<span class="tbartitle">zsh — ' + esc(m.cwd || 'workspace') + '</span></div>'
+      + '<div class="termbody">'
+        + '<div class="termline"><span class="tprompt">➜</span> <span class="tcwd">' + esc(m.cwd || '') + '</span> '
+        + '<span class="tcmd">' + esc(m.command || '') + '</span></div>'
+        + '<pre class="termout"></pre>'
+        + '<div class="termstatus running"><span class="tcaret"></span> running… '
+        + '<button class="termstop" title="Stop this command">' + codicon('stop-circle') + ' Stop</button></div>'
+      + '</div>';
+    log.appendChild(card); scrollIfStuck();
+    termCards[m.id] = card;
+    const sb = card.querySelector('.termstop'); if (sb) { sb.onclick = () => { sb.disabled = true; sb.textContent = 'stopping…'; vscode.postMessage({ type: 'stopCommand', id: m.id }); }; }
+  }
+  function termOutputAppend(m){
+    const card = termCards[m.id]; if (!card) { return; }
+    const out = card.querySelector('.termout'); if (!out) { return; }
+    out.insertAdjacentHTML('beforeend', m.stream === 'stderr' ? '<span class="termerr">' + esc(m.chunk) + '</span>' : esc(m.chunk));
+    out.scrollTop = out.scrollHeight;   // keep the latest line visible inside the card
+    scrollIfStuck();
+  }
+  function termExitFinish(m){
+    const card = termCards[m.id]; if (!card) { return; }
+    const st = card.querySelector('.termstatus'); if (!st) { return; }
+    const dur = (typeof m.ms === 'number') ? (m.ms >= 1000 ? (m.ms / 1000).toFixed(1) + 's' : m.ms + 'ms') : '';
+    if (m.how === 'stopped'){ st.innerHTML = codicon('circle-slash') + ' stopped' + (dur ? ' · ' + dur : ''); st.className = 'termstatus bad'; }
+    else if (m.how === 'timeout'){ st.innerHTML = codicon('warning') + ' timed out' + (dur ? ' · ' + dur : ''); st.className = 'termstatus bad'; }
+    else { const ok = m.code === 0; st.innerHTML = (ok ? codicon('check') : codicon('circle-slash')) + ' exit ' + m.code + (dur ? ' · ' + dur : ''); st.className = 'termstatus ' + (ok ? 'ok' : 'bad'); }
+    delete termCards[m.id];
+    scrollIfStuck();
+  }
+
   // Blocking clarifying questions (ask_user) — clickable numbered options the agent waits on.
   function addQuestions(m){
     clearEmpty(); clearStatus(); agentBubble = null;
@@ -1253,6 +1305,9 @@
     else if (m.type === 'agentTool'){ finishAgentBubble(); addAgentLine(m.icon, m.text); }
     else if (m.type === 'agentApproval'){ addApproval(m); }
     else if (m.type === 'agentQuestions'){ addQuestions(m); }
+    else if (m.type === 'termRun'){ addTermRun(m); }
+    else if (m.type === 'termOutput'){ termOutputAppend(m); }
+    else if (m.type === 'termExit'){ termExitFinish(m); }
     else if (m.type === 'editApplied'){ addEditCard(m); }
     else if (m.type === 'editResolved'){ resolveEditCard(m); }
     else if (m.type === 'reviewState'){ updateReviewBar(m.count, m.add, m.del); }
@@ -1268,7 +1323,7 @@
     else if (m.type === 'reset'){
       log.innerHTML = '<div id="empty"><div class="big">New chat</div>Ask about your code, or describe what to build.</div>';
       selLabel = null; ctxFiles = []; renderChips();
-      for (const k in editCards) { delete editCards[k]; } updateReviewBar(0); renderPlan([]); renderContext({ input: 0 });
+      for (const k in editCards) { delete editCards[k]; } for (const k in termCards) { delete termCards[k]; } updateReviewBar(0); renderPlan([]); renderContext({ input: 0 });
       pending = ''; shown = ''; doneSignaled = false; flushAll = false; current = null; setStreaming(false);
       forceStick();
     }
 
@@ -120,6 +120,11 @@
 					"default": "",
 					"markdownDescription": "Base URL of your Atom++ Cloud account server (for sign-in + settings/skills sync). Leave empty until the backend exists (roadmap M9). **Your AI provider keys never go to this server — AI requests stay direct from your machine to your provider.**"
 				},
+				"atompp.ai.commandTimeout": {
+					"type": "number",
+					"default": 120000,
+					"description": "Backstop timeout in milliseconds before an approved shell command is force-killed (its whole process group). The Stop button on the terminal card is the primary control; set 0 to disable the backstop so long-running servers run until you stop them."
+				},
 				"atompp.ai.contextWindow": {
 					"type": "number",
 					"default": 200000,