@noobdemon/noob-cli 1.12.16 → 1.12.18

package/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 Tất cả thay đổi đáng kể của `@noobdemon/noob-cli` được ghi vào file này.
 
+## [1.12.18] - 2026-06-25
+
+### Changed
+- Cải thiện system prompt cho coding flow: inspect đúng chỗ, patch hẹp, verify nhanh và final gọn hơn để model code mượt, ít vòng lặp hơn.
+
+## [1.12.17] - 2026-06-24
+
+### Added
+- **Image input support** (`src/tui.js`, `src/repl.js`, `src/api.js`, `src/agent.js`, `worker/src/worker.js`): rich TTY trên Windows hỗ trợ `Alt+V` để lấy ảnh từ clipboard, hiện chip `[pasted image #1]`, rồi gửi kèm payload `image: "data:image/png;base64,..."` qua gateway. CLI cũng tự đính kèm ảnh đầu tiên khi user nhắc `@file.png|jpg|jpeg|webp|gif` (giới hạn 8MB). Worker validate data URL và forward `image` sang Railway (`{message, model, image}`); fallback cũ giữ `{message, model}`.
+
+### Verified
+- `npm test` 100/100 pass.
+- Railway `/chat` smoke với `image` data URL base64 trả mô tả ảnh thành công.
+
 ## [1.12.15] - 2026-06-16
 
 ### Fixed

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@noobdemon/noob-cli",
-  "version": "1.12.16",
+  "version": "1.12.18",
   "publishConfig": {
     "access": "public"
   },

package/src/agent.js

@@ -36,7 +36,7 @@ Available tools (each is self-contained; pick the SMALLEST tool that answers the
 Context is finite. Don't slurp the whole repo up front. Discover information progressively: list_dir/glob to map → grep to locate → read_file (with offset+limit for big files) to inspect only what matters. Each tool result spends your attention budget — make every call earn it. When a tool returns a huge blob, extract the few facts you need, then move on; don't re-read it later (the result stays in history).
 
 # Rules
-- TODO-BASED EXECUTION: For any multi-step task (3+ actions), you MUST call \`write_todos\` FIRST with all items done:false, then call it AGAIN after every completed step with that item flipped to done:true (resend the full list). NEVER write markdown \`- [ ]\` lines — the runtime parses \`write_todos\` calls, not markdown. Your response is NOT finished until all items are done:true. The ONLY valid reason to stop is: (a) all items done, or (b) you are WAITING for a user reply. If you just got a tool result, you MUST continue — do NOT output a summary, do NOT ask "what next", do NOT stop. After write_file/edit_file returns, call write_todos to tick the just-finished item, then immediately start the next.
+- TODO-BASED EXECUTION: For any multi-step task (3+ actions), you MUST call \`write_todos\` FIRST with all items done:false, then call it AGAIN after every completed step with that item flipped to done:true (resend the full list). NEVER write markdown \`- [ ]\` lines — the runtime parses \`write_todos\` calls, not markdown. Your response is NOT finished until all items are done:true. The ONLY valid reason to stop is: (a) all items done and sensible verification is complete, or (b) you are WAITING for a user reply. If you just got a tool result and unfinished work remains, continue with the next tool — do NOT output a premature summary, do NOT ask "what next". After write_file/edit_file returns, call write_todos to tick the just-finished item, then immediately start the next.
 - GROUND TRUTH = real TOOL RESULTs in this conversation, not your memory or what you intended to do. A file changed only if a write_file/edit_file result confirms it (see the FILES CHANGED list). A test passed / build succeeded / command worked only if a run_command result above shows it. Never narrate outcomes you didn't observe; if you haven't checked, say so and check now (read_file / list_dir / run the command). Before any "done/summary" reply, reconcile every file and result you're about to claim against the actual tool results above — if it isn't there, you didn't do it yet.
 - VERIFY BEFORE DISMISSING: never declare a TOOL RESULT "fake", "spurious", "injected", "unrelated", or "from a previous turn" without first verifying with a fresh tool call. If a result looks off (unexpected content, output you didn't ask for, weird command), your DEFAULT is: treat it as REAL runtime output, then run a small verification (read_file the affected path, grep for the symbol, list_dir, re-run the command) to confirm actual state. Only after the verification tool result contradicts the suspicious one may you call it stale/leftover — and even then, work from the FRESH result, never from your guess. Trusting your own skepticism over the runtime is the same over-confidence bug as hallucinating success: both substitute memory for evidence.
 - Investigate before editing: read the relevant files first; never invent file contents.
@@ -66,6 +66,12 @@ Context is finite. Don't slurp the whole repo up front. Discover information pro
 3. SURGICAL: change only what the task needs. No drive-by refactors, renames, reformatting, or comment churn in unrelated code.
 4. VERIFIABLE GOAL: decide how you'll know it works, then check it (run the build/test, read the output). Report what you verified — and honestly state what you did NOT verify.
 
+# Coding workflow — default for implementation tasks
+1. Inspect first: list/glob/grep/read only the files needed to understand the change.
+2. Patch narrowly: edit the smallest relevant block; avoid new helpers/files unless they clearly reduce code now.
+3. Verify narrowly: run the fastest relevant test/lint/build command. If it fails, use the output as ground truth and fix once before broadening scope.
+4. Finish cleanly: final answer states the changed files and the exact verification result. Do not include tool-call JSON, long plans, or speculative next steps.
+
 # Example interaction
 ## USER
 do the tests pass?
@@ -556,6 +562,7 @@ function extractJsonObject(s, from) {
 export async function runAgent({
   history,
   model,
+  image,
   signal,
   onTool,
   onStatus,
@@ -608,6 +615,7 @@ export async function runAgent({
     const { text, finishReason } = await streamWithRetry({
       model,
       message,
+      image,
       system,
       signal,
       tokenMeter,
@@ -756,6 +764,7 @@ export async function runAgent({
 async function streamWithRetry({
   model,
   message,
+  image,
   system,
   signal,
   tokenMeter,
@@ -771,6 +780,7 @@ async function streamWithRetry({
         mode: 'chat',
         model,
         message,
+        image,
         system,
         signal,
         effort,

package/src/api.js

@@ -157,6 +157,7 @@ function hasUnclosedToolBlock(text) {
 export async function stream({
   mode = 'chat',
   message,
+  image,
   model,
   system,
   conversation,
@@ -186,6 +187,7 @@ export async function stream({
       endpoint,
       mode,
       message: prompt,
+      image,
       model,
       system,
       conversation,
@@ -262,6 +264,7 @@ async function streamOnce({
   endpoint,
   mode,
   message,
+  image,
   model,
   system,
   conversation,
@@ -278,6 +281,7 @@ async function streamOnce({
   else if (mode === 'merge') body = { message };
   else {
     body = { message, model, remember: true, memoryToken: getMemoryToken() };
+    if (image) body.image = image;
     if (system) body.customInstructions = system;
     if (Array.isArray(conversation) && conversation.length) body.conversation = conversation;
     if (effort) body.effort = effort;

package/src/repl.js

@@ -101,6 +101,24 @@ import {
 } from './repl/utils.js';
 import { createAgentDispatcher } from './repl/agent-dispatch.js';
 import { createBgRegistry } from './workflow-bg.js';
+
+const IMAGE_MIME = {
+  '.png': 'image/png',
+  '.jpg': 'image/jpeg',
+  '.jpeg': 'image/jpeg',
+  '.webp': 'image/webp',
+  '.gif': 'image/gif',
+};
+
+function imageDataUrl(file) {
+  if (!file) return null;
+  const mime = IMAGE_MIME[path.extname(file).toLowerCase()];
+  if (!mime) return null;
+  const abs = path.resolve(process.cwd(), file);
+  const stat = fs.statSync(abs);
+  if (stat.size > 8 * 1024 * 1024) throw new Error('Ảnh quá lớn (>8MB): ' + file);
+  return `data:${mime};base64,${fs.readFileSync(abs).toString('base64')}`;
+}
 export async function startRepl(opts = {}) {
   const state = createState(opts, config);
   const tokenMeter = new TokenMeter();
@@ -1453,23 +1471,25 @@ NGUYÊN TẮC:
     let input;
     if (pending.length) {
       // Có tin xếp hàng → tự gửi câu kế tiếp (không chờ gõ).
-      input = (pending.shift() ?? '').trim();
+      input = { text: (pending.shift() ?? '').trim(), images: [] };
     } else {
       const raw = await ask(promptStr(false));
       if (raw == null) break; // stdin fully closed and drained
-      input = raw.trim();
+      input = typeof raw === 'string' ? { text: raw.trim(), images: [] } : raw;
+      input.text = String(input.text || '').trim();
+      input.images = Array.isArray(input.images) ? input.images : [];
     }
-    if (!input) continue;
+    if (!input.text) continue;
     // Bọc cả lượt: một lỗi trong xử lý lệnh/agent không được phép thoát ra
     // ngoài vòng lặp (sẽ rơi vào .catch ở bin/noob.js → process.exit(1) =
     // "tự động tắt"). Bắt ở đây, in lỗi, rồi tiếp tục vòng lặp.
     try {
-      if (input.startsWith('/')) {
-        const done = await command(input);
+      if (input.text.startsWith('/')) {
+        const done = await command(input.text);
         if (done) break;
         continue;
       }
-      await handle(input);
+      await handle(input.text, { images: input.images });
       persist(); // lưu sau mỗi lượt → resume được kể cả khi tắt đột ngột
     } catch (err) {
       printError(err);
@@ -1481,7 +1501,7 @@ NGUYÊN TẮC:
   process.exit(0);
 
   // ── turn handler ─────────────────────────────────────────────────────────
-  async function handle(text) {
+  async function handle(text, opts = {}) {
     if (!config.apiKey) {
       console.log(c.tool('  ' + t.notLoggedIn));
       return;
@@ -1546,9 +1566,12 @@ NGUYÊN TẮC:
       }
 
       const files = mentionedFiles(text);
+      const image =
+        opts.images?.[0] ||
+        imageDataUrl(files.find((f) => IMAGE_MIME[path.extname(f).toLowerCase()]));
       const content = files.length
         ? text +
-          `\n\n[File người dùng nhắc tới bằng @: ${files.join(', ')} — đọc bằng read_file nếu cần.]`
+          `\n\n[File người dùng nhắc tới bằng @: ${files.join(', ')} — đọc bằng read_file nếu cần.${image ? ' Ảnh đầu tiên đã được đính kèm cho model vision.' : ''}]`
         : text;
       state.history.push({ role: 'user', content });
       // Update terminal title với session name (trích từ message đầu).
@@ -1592,6 +1615,7 @@ NGUYÊN TẮC:
       const answer = await runAgent({
         history: state.history,
         model: state.model.id,
+        image,
         signal: abort.signal,
         tokenMeter,
         goal: state.goal,

package/src/tui.js

@@ -7,11 +7,32 @@
 // Bật/tắt: TTY thật → chế độ giàu; không phải TTY hoặc NOOB_TUI=0 → chế độ "dumb"
 // (đọc dòng đơn giản, in thẳng) để khỏi vỡ ở terminal lạ / pipe / CI.
 import readline from 'node:readline';
+import { execFileSync } from 'node:child_process';
 import { c } from './ui.js';
 
 const ESC = '\x1b';
 const ANSI_RE = /\x1b\[[0-9;?]*[ -/]*[@-~]/g;
 const visLen = (s) => s.replace(ANSI_RE, '').length;
+function readClipboardImageDataUrl() {
+  if (process.platform !== 'win32') return null;
+  try {
+    const script =
+      'Add-Type -AssemblyName System.Windows.Forms,System.Drawing;' +
+      '$img=[Windows.Forms.Clipboard]::GetImage();' +
+      'if($null -eq $img){exit 2};' +
+      '$ms=New-Object IO.MemoryStream;' +
+      '$img.Save($ms,[Drawing.Imaging.ImageFormat]::Png);' +
+      '[Convert]::ToBase64String($ms.ToArray())';
+    const b64 = execFileSync('powershell.exe', ['-NoProfile', '-STA', '-Command', script], {
+      encoding: 'utf8',
+      timeout: 5000,
+      windowsHide: true,
+    }).trim();
+    return b64 ? `data:image/png;base64,${b64}` : null;
+  } catch {
+    return null;
+  }
+}
 // Trả index trong `text` mà tại đó vị trí VISUAL đạt `targetVis`. Bỏ qua toàn
 // bộ ANSI escape sequence khi đếm. Dùng cho soft-wrap khi text có màu.
 function findVisPos(text, targetVis) {
@@ -207,6 +228,7 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
   // vị trí con trỏ → hỗ trợ ←/→/Home/End/Delete, chèn & xoá GIỮA dòng.
   let cells = [];
   let cur = 0;
+  let imageSeq = 0;
   let waiter = null;
   const queue = [];
 
@@ -254,8 +276,13 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
       ? firstLine.slice(0, PASTE_PREVIEW_MAX - 1) + '…'
       : firstLine;
   };
-  const cellStr = (x) => (x.paste !== undefined ? x.paste : x.c);
+  const imageLabel = (x) => `[pasted image #${x.n}]`;
+  const cellStr = (x) => {
+    if (x.image) return imageLabel(x);
+    return x.paste !== undefined ? x.paste : x.c;
+  };
   const cellPlain = (x) => {
+    if (x.image) return imageLabel(x);
     if (x.paste === undefined) return x.c;
     const preview = pastePreview(x.paste);
     return preview ? `[pasted ${x.lines} lines: "${preview}"]` : `[pasted ${x.lines} lines]`;
@@ -264,6 +291,7 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
   const coloredInput = () =>
     cells
       .map((x) => {
+        if (x.image) return c.dim(imageLabel(x));
         if (x.paste === undefined) return x.c;
         const preview = pastePreview(x.paste);
         const label = preview
@@ -568,6 +596,18 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
       histPos = null;
     } else pushText(content); // paste 1 dòng = gõ thẳng
   }
+  function pushImage(image) {
+    imageSeq += 1;
+    cells.splice(cur, 0, { image, n: imageSeq });
+    cur += 1;
+    histPos = null;
+  }
+  function pasteClipboardImage() {
+    const image = readClipboardImageDataUrl();
+    if (image) pushImage(image);
+    else w('\x07');
+    draw();
+  }
   function backspace() {
     if (cur > 0) {
       cells.splice(cur - 1, 1);
@@ -582,7 +622,8 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
     }
   }
   // null = ô dán (coi như ranh giới từ); ngược lại trả ký tự của ô.
-  const charAt = (k) => (cells[k] && cells[k].paste === undefined ? cells[k].c : null);
+  const charAt = (k) =>
+    cells[k] && cells[k].paste === undefined && !cells[k].image ? cells[k].c : null;
   function moveWordLeft() {
     while (cur > 0 && charAt(cur - 1) === ' ') cur -= 1;
     if (cur > 0 && charAt(cur - 1) === null) return void (cur -= 1); // qua 1 chip
@@ -639,6 +680,7 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
 
   function submit() {
     const full = fullText();
+    const images = cells.filter((x) => x.image).map((x) => x.image);
     const echo = echoUserBlock();
     if (full.trim() && submitHistory[submitHistory.length - 1] !== full) {
       submitHistory.push(full);
@@ -654,9 +696,9 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
     if (waiter) {
       const wr = waiter;
       waiter = null;
-      wr(full);
-    } else if (onLine) onLine(full);
-    else queue.push(full);
+      wr(images.length ? { text: full, images } : full);
+    } else if (onLine) onLine(images.length ? { text: full, images } : full);
+    else queue.push(images.length ? { text: full, images } : full);
   }
 
   let carry = '';
@@ -702,6 +744,11 @@ export function createTui({ onLine, onInterrupt, onEOF, onShiftTab, onCtrlT, com
       }
       const ch = s[i];
       if (ch === ESC) {
+        if (rest[1] === 'v' || rest[1] === 'V') {
+          pasteClipboardImage();
+          i += 2;
+          continue;
+        }
         // Một chuỗi CSI hoàn chỉnh: \x1b[ … <chữ/~>  hoặc SS3: \x1bO<chữ>.
         const csi = rest.match(/^\x1b\[[0-9;]*[~A-Za-z]/) || rest.match(/^\x1bO[A-Za-z]/);
         if (!csi) {