npm - coxpit - Versions diffs - 2.8.0 → 2.9.0 - Mend

coxpit 2.8.0 → 2.9.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (4) hide show

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "coxpit",
-  "version": "2.8.0",
+  "version": "2.9.0",
   "description": "Self-hosted cockpit for running a fleet of AI coding agents across your own machines — parallel worktree runs, live board, compare & merge, web terminal, design capture.",
   "type": "module",
   "license": "MIT",

package/src/board.ts CHANGED Viewed

@@ -263,6 +263,18 @@ export const BOARD_HTML = /* html */ `<!doctype html>
   .cmp-f .msg{font-family:var(--mono);font-size:11px;color:var(--muted);flex:1;
     white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
+  /* ── AI review panel (compare) ── */
+  .cmp-review{border-bottom:1px solid var(--line);background:var(--surface2);padding:14px 20px;
+    max-height:42vh;overflow:auto;font-size:13px;line-height:1.65;color:var(--muted)}
+  .cmp-review[hidden]{display:none}
+  .cmp-review h2{font-size:13px;color:var(--brand);margin:14px 0 6px;letter-spacing:.02em}
+  .cmp-review h3{font-size:12.5px;color:var(--ink);margin:12px 0 4px}
+  .cmp-review ul{margin:4px 0 8px;padding-left:18px}
+  .cmp-review li{margin-bottom:3px}
+  .cmp-review strong{color:var(--ink)}
+  .cmp-review code{font-family:var(--mono);font-size:.9em;background:#0e1118;padding:1px 5px;border-radius:4px;color:var(--brand)}
+  .cmp-review p{margin:0 0 8px}
   /* ── terminal ───────────────────────────── */
   .term-body{flex:1;min-height:0;background:#0b0d12;padding:8px 4px 4px 10px}
   #xterm{width:100%;height:100%}
@@ -390,9 +402,11 @@ export const BOARD_HTML = /* html */ `<!doctype html>
   <div class="modal wide">
     <div class="modal-h">
       <span class="title" id="cmpTitle">Compare</span>
+      <button class="btn sm" id="cmpAI">AI review</button>
       <button class="btn-ghost sm" id="cmpRefresh">Refresh</button>
       <button class="x" id="cmpClose" aria-label="close">×</button>
     </div>
+    <div class="cmp-review" id="cmpReview" hidden></div>
     <div class="cmp" id="cmpBody"></div>
   </div>
 </div>
@@ -578,20 +592,49 @@ $('cfmOk').addEventListener('click', ()=>cfmClose(true));
 $('cfmCancel').addEventListener('click', ()=>cfmClose(false));
 $('cfmOverlay').addEventListener('click',(e)=>{ if(e.target===$('cfmOverlay')) cfmClose(false); });
-function summarize(kind, payload){
+/* 이벤트 인간화 — JSON 원문 대신 사람이 읽는 한 줄로. null = 표시 생략(노이즈). */
+function humanize(e){
+  const kind = e.kind, payload = e.payload;
+  if (kind === 'rate_limit_event') return null;
+  if (kind === 'steer') return { k:'steer', t:'→ '+payload };
+  if (kind === 'export'){ try{ const o=JSON.parse(payload); return { k:'export', t:o.copied+' file(s) → '+o.dest }; }catch{ return { k:'export', t:payload }; } }
+  if (kind === 'pr') return { k:'pr', t:payload };
+  if (kind === 'stderr') return { k:'stderr', t:payload };
   try{
     const o = JSON.parse(payload);
+    if (o.type === 'system') return { k:'session', t:'started · '+(o.model||o.subtype||'') };
+    if (o.type === 'user') return null; // tool 결과 회신 — 노이즈
     if (o.type === 'assistant' && o.message){
-      const c = (o.message.content||[]).map(x => x.type==='text' ? x.text : (x.type==='tool_use' ? '['+x.name+']' : '')).join(' ');
-      return c || '(assistant)';
+      const parts = [];
+      for (const x of (o.message.content||[])){
+        if (x.type === 'text' && x.text) parts.push({ k:'said', t:x.text });
+        else if (x.type === 'tool_use'){
+          const i = x.input || {};
+          const arg = i.file_path || i.command || i.path || i.pattern || '';
+          parts.push({ k:'tool', t:'▸ '+x.name+(arg?' — '+String(arg).split('/').slice(-2).join('/').slice(0,60):'') });
+        }
+      }
+      return parts.length ? parts : null;
     }
-    if (o.type === 'assistant' && o.text) return o.text;
-    if (o.type === 'result') return o.result || '(result)';
-    if (o.type === 'user') return '(tool result)';
-    if (o.type === 'system') return o.subtype || 'system';
-    if (kind === 'meta') return 'worktree ' + (o.worktree||'');
-    return kind;
-  }catch{ return payload; }
+    if (o.type === 'assistant' && o.text) return { k:'said', t:o.text };
+    if (o.type === 'result') return { k:'done', t:o.result || 'finished' };
+    if (kind === 'meta') return { k:'start', t:'worktree '+String(o.worktree||'').split('/').slice(-2).join('/') };
+    return { k:kind, t:payload.slice(0,140) };
+  }catch{ return { k:kind, t:payload }; }
+}
+function humanLines(events){
+  const out = [];
+  for (const e of (events||[])){
+    const h = humanize(e);
+    if (!h) continue;
+    if (Array.isArray(h)) out.push(...h); else out.push(h);
+  }
+  return out;
+}
+function summarize(kind, payload){
+  const h = humanize({kind, payload});
+  if (!h) return '';
+  return Array.isArray(h) ? h.map(x=>x.t).join(' · ') : h.t;
 }
 function diffHTML(text){
   if (!text.trim()) return '<span style="color:var(--faint)">no changes</span>';
@@ -667,8 +710,8 @@ function cardHTML(r){
   const closed = task && task.status==='closed';
   const title = (task ? esc(task.title) : ('task ' + (r.taskId ?? '?')))
     + (closed ? ' <span class="closed">· closed</span>' : '');
-  const evs = (r.events||[]).slice(-8).map(e =>
-    '<div class="ev"><span class="k">'+esc(e.kind)+'</span><span class="t">'+esc(summarize(e.kind,e.payload)).slice(0,140)+'</span></div>'
+  const evs = humanLines(r.events).slice(-8).map(h =>
+    '<div class="ev"><span class="k">'+esc(h.k)+'</span><span class="t">'+esc(h.t).slice(0,140)+'</span></div>'
   ).join('') || '<div class="ev"><span class="t" style="color:var(--faint)">waiting…</span></div>';
   const selCls = (selectMode?' selmode':'') + (selected.has(r.id)?' selected':'');
   return '<div class="card'+selCls+'" id="card-'+r.id+'">'
@@ -787,8 +830,8 @@ function paintModal(){
   $('mStop').style.display = (r.status==='running'||r.status==='preparing'||r.status==='pending') ? '' : 'none';
   // steer 는 정착한 real run 에서만 의미(드라이런은 세션 없음 — 서버가 사유와 함께 거절)
   $('steerRow').style.display = ['done','failed','stopped'].includes(r.status) ? '' : 'none';
-  $('mTimeline').innerHTML = (r.events||[]).map(e =>
-    '<div class="ev"><span class="k">'+esc(e.kind)+'</span><span class="t">'+esc(summarize(e.kind,e.payload))+'</span></div>'
+  $('mTimeline').innerHTML = humanLines(r.events).map(h =>
+    '<div class="ev"><span class="k">'+esc(h.k)+'</span><span class="t">'+esc(h.t)+'</span></div>'
   ).join('') || '<span style="color:var(--faint)">no events yet</span>';
 }
 async function loadDiff(){
@@ -908,6 +951,7 @@ $('mCloseTask').addEventListener('click', async ()=>{
 let cmpTaskId = null;
 async function openCompare(taskId){
   cmpTaskId = taskId;
+  $('cmpReview').hidden = true; $('cmpReview').innerHTML = '';
   $('cmpOverlay').classList.add('open');
   $('cmpBody').innerHTML = '<div class="empty" style="flex:1">loading…</div>';
   await paintCompare();
@@ -969,6 +1013,34 @@ $('cmpBody').addEventListener('click', async (e)=>{
 $('cmpClose').addEventListener('click', ()=>{ cmpTaskId=null; $('cmpOverlay').classList.remove('open'); });
 $('cmpOverlay').addEventListener('click',(e)=>{ if(e.target===$('cmpOverlay')){ cmpTaskId=null; $('cmpOverlay').classList.remove('open'); } });
 $('cmpRefresh').addEventListener('click', paintCompare);
+/* 초경량 md 렌더 (리뷰 표시용) */
+function mdLite(src){
+  let s = esc(src);
+  s = s.replace(/\`\`\`[a-z]*\\n([\\s\\S]*?)\`\`\`/g, (m,c)=>'<pre style="background:#0e1118;border:1px solid var(--line);border-radius:7px;padding:8px 10px;overflow-x:auto">'+c+'</pre>');
+  s = s.replace(/^### (.+)$/gm, '<h3>$1</h3>');
+  s = s.replace(/^## (.+)$/gm, '<h2>$1</h2>');
+  s = s.replace(/\\*\\*([^*]+)\\*\\*/g, '<strong>$1</strong>');
+  s = s.replace(/\`([^\`]+)\`/g, '<code>$1</code>');
+  s = s.replace(/^[-*] (.+)$/gm, '<li>$1</li>');
+  s = s.replace(/(<li>[\\s\\S]*?<\\/li>)(?!\\s*<li>)/g, '<ul>$1</ul>');
+  s = s.split(/\\n{2,}/).map(b => /^<(h2|h3|ul|pre)/.test(b.trim()) ? b : (b.trim()? '<p>'+b.replace(/\\n/g,'<br>')+'</p>':'' )).join('');
+  return s;
+}
+$('cmpAI').addEventListener('click', async ()=>{
+  if (cmpTaskId==null) return;
+  const yes = await confirmUI('Run an AI review of these implementations?',
+    { sub: 'A reviewer agent reads every diff and summarizes each approach, pros/cons, and a recommendation — so you judge instead of reading all the code. Real agent, spends credits (~1–2 min).', okLabel: 'Review' });
+  if (!yes) return;
+  const btn = $('cmpAI');
+  btn.disabled = true; btn.textContent = 'Reviewing…';
+  try{
+    const res = await fetch('/api/tasks/'+cmpTaskId+'/review',{method:'POST',
+      headers:{'content-type':'application/json'}, body:JSON.stringify({real:true})});
+    const j = await res.json().catch(()=>({}));
+    if (res.ok){ $('cmpReview').innerHTML = mdLite(j.review||''); $('cmpReview').hidden = false; }
+    else toast('review: '+(j.detail||res.status), 'error');
+  } finally { btn.disabled = false; btn.textContent = 'AI review'; }
+});
 $('mCompare').addEventListener('click', ()=>{
   if (openRunId==null) return;
   const r = runs.get(openRunId); if(!r) return;

package/src/orchestrator.ts CHANGED Viewed

@@ -400,6 +400,62 @@ export async function planFanout(repoId: number, goal: string, real: boolean): P
   return { ok: true, detail: `${created.length} task(s) launched`, tasks: created };
 }
+/**
+ * AI 리뷰(심판) — 태스크의 정착 run diff 들을 리뷰 에이전트가 읽고
+ * 접근 방식·장단점·리스크·추천을 요약한다. 사람은 코드 전수가 아니라
+ * 판단만 하면 되도록. (read-only, 워크트리 불필요)
+ */
+export async function reviewTask(taskId: number, real: boolean): Promise<{ ok: boolean; detail: string; review?: string }> {
+  const tr = await db.select().from(tasks).where(eq(tasks.id, taskId)).limit(1);
+  const task = tr[0];
+  if (!task) return { ok: false, detail: 'task not found' };
+  const rp = await db.select().from(repos).where(eq(repos.id, task.repoId)).limit(1);
+  const repo = rp[0];
+  if (!repo) return { ok: false, detail: 'repo not found' };
+  const mr = await db.select().from(machines).where(eq(machines.id, repo.machineId)).limit(1);
+  const m = mr[0];
+  if (!m) return { ok: false, detail: 'machine not found' };
+  const machine: MachineTarget = { slug: m.slug, kind: m.kind, address: m.address, sshUser: m.sshUser };
+  const trs = (await db.select().from(agentRuns).where(eq(agentRuns.taskId, taskId)))
+    .filter((r) => ['done', 'failed', 'stopped', 'merged'].includes(r.status));
+  if (trs.length < 2) return { ok: false, detail: 'need at least 2 settled runs to review' };
+  const sections: string[] = [];
+  for (const r of trs) {
+    const d = await getRunDiff(r.id);
+    const diff = (d.ok ? d.diff : '(worktree gone — diff unavailable)').slice(0, 15000);
+    sections.push(`### run r${r.id} (status: ${r.status})\nAgent's own summary: ${(r.exitSummary || '-').slice(0, 300)}\n\nDiff:\n\`\`\`diff\n${diff}\n\`\`\``);
+  }
+  if (!real) {
+    return {
+      ok: true, detail: 'rehearsal review',
+      review: `## AI Review (rehearsal)\n\n${trs.map((r) => `**r${r.id}** — approach: (dry-run placeholder)\n- pros: n/a\n- cons: n/a`).join('\n\n')}\n\n**Recommendation**: run with Real agent for an actual review.`,
+    };
+  }
+  const prompt =
+    `You are reviewing ${trs.length} competing implementations of the same task.\n` +
+    `Task: ${task.title}\nOriginal prompt: ${task.prompt.slice(0, 800)}\n\n` +
+    sections.join('\n\n') +
+    `\n\nWrite a review in markdown, in the language of the task prompt (Korean if the prompt is Korean):\n` +
+    `1. For EACH run: one-line approach summary, then pros (max 3) and cons (max 3) as bullets.\n` +
+    `2. '## 추천' section: which run to merge and WHY, in 2-3 sentences. If combining both is better, say exactly what to steer.\n` +
+    `Judge correctness, simplicity, consistency with the existing codebase, and risk. Be decisive. Respond with ONLY the markdown.`;
+  const cmd = `cd ${shq(repo.path)} && ${config.agent.bin} -p ${shq(prompt)} --output-format json`;
+  const r = await runShellOn(machine, cmd, 300000);
+  if (!r.ok) return { ok: false, detail: 'reviewer failed: ' + (r.stderr || r.stdout).trim().slice(0, 300) };
+  try {
+    const envelope = JSON.parse(r.stdout.trim()) as { result?: string };
+    const review = (envelope.result ?? '').trim();
+    if (!review) throw new Error('empty review');
+    return { ok: true, detail: 'reviewed', review };
+  } catch (e) {
+    return { ok: false, detail: 'could not parse review: ' + String(e).slice(0, 200) };
+  }
+}
 export interface IntegrateResult {
   runId: number;
   status: 'merged' | 'conflict' | 'skipped';

package/src/server.ts CHANGED Viewed

@@ -12,7 +12,7 @@ import { db } from './db';
 import { machines, repos, tasks, agentRuns, agentEvents, designCaptures } from './db/schema';
 import { BOOKMARKLET_JS } from './design';
 import { runShellOn, shq } from './exec';
-import { launchRun, cleanupRun, stopRun, getRunDiff, mergeRun, getRunTermInfo, steerRun, exportRun, prRun, integrateRuns, planFanout } from './orchestrator';
+import { launchRun, cleanupRun, stopRun, getRunDiff, mergeRun, getRunTermInfo, steerRun, exportRun, prRun, integrateRuns, planFanout, reviewTask } from './orchestrator';
 import { openTerm } from './term';
 import { addSink, removeSink, broadcast } from './hub';
 import { BOARD_HTML } from './board';
@@ -32,7 +32,7 @@ export async function buildServer(): Promise<FastifyInstance> {
   app.addHook('onRequest', authGate);
   // 무인증 헬스(외부 감시용)
-  app.get('/api/health', async () => ({ ok: true, name: 'coxpit', version: '2.8.0' }));
+  app.get('/api/health', async () => ({ ok: true, name: 'coxpit', version: '2.9.0' }));
   // 플릿 보드(단일 페이지). 인증 게이트 적용됨.
   app.get('/', async (_req, reply) => reply.type('text/html').send(BOARD_HTML));
@@ -325,6 +325,17 @@ export async function buildServer(): Promise<FastifyInstance> {
     return reply.code(202).send({ ok: true, runs: created.map((r) => ({ id: r.id, status: r.status })) });
   });
+  // AI 리뷰 — 심판 에이전트가 run diff 들을 읽고 접근/장단점/추천을 요약.
+  app.post('/api/tasks/:id/review', async (req, reply) => {
+    const id = Number((req.params as { id: string }).id);
+    const b = (req.body ?? {}) as { real?: boolean };
+    const tr = await db.select().from(tasks).where(eq(tasks.id, id)).limit(1);
+    if (!tr[0]) return reply.code(404).send({ error: 'task not found' });
+    const res = await reviewTask(id, b.real === true);
+    if (!res.ok) return reply.code(422).send(res);
+    return res;
+  });
   // 비교 뷰 — 태스크의 모든 run + 각 diff 를 한 방에 (승자 고르기용).
   app.get('/api/tasks/:id/compare', async (req, reply) => {
     const id = Number((req.params as { id: string }).id);
@@ -470,7 +481,7 @@ export async function buildServer(): Promise<FastifyInstance> {
   // 라이브 스트림 좌석 — 오케스트레이터가 run/event 를 여기로 broadcast.
   app.get('/ws', { websocket: true }, (socket) => {
     addSink(socket);
-    socket.send(JSON.stringify({ type: 'hello', name: 'coxpit-fleet', version: '2.8.0' }));
+    socket.send(JSON.stringify({ type: 'hello', name: 'coxpit-fleet', version: '2.9.0' }));
     socket.on('close', () => removeSink(socket));
   });