From 128f0f3694ae4ddeb5833c94a7c40ac9e22b4c31 Mon Sep 17 00:00:00 2001
From: Agando <roduccomm@gmail.com>
Date: Wed, 21 Oct 2026 22:50:03 +0200
Subject: fix(newos4): GLM review follow-ups for AI read/context (v4.6.47)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- loop guard: a mutation unblocks the replay but no longer erases the repeat counter
  (identical mutating calls could loop forever)
- tool traces are per agent tag, flushed in a catch as well as on success, ring-buffered
  (first 2 + last 10) so verification reads survive
- read_file notes are metadata only (path/range/next_offset); no file bodies, no secrets,
  and web_search/web_fetch output is excluded from durable context (injection surface)
- trace is role:'system' (was 'assistant' — the model re-read it as its own answer)
- surrogate-safe pagination, 500 floor / 16000 cap, prompt limits aligned
- prompt now names checks that exist in the in-page terminal (test, grep -n, wc -l)
- compressed summary capped at 4000 chars; denied-destructive tool result carries tool_call_id
---
 main.html                       |  6 +--
 modules/core/core-bundle.js     | 93 +++++++++++++++++++++++++++++++----------
 modules/core/module-aiassist.js | 93 +++++++++++++++++++++++++++++++----------
 modules/core/module-loader.js   |  2 +-
 4 files changed, 144 insertions(+), 50 deletions(-)

diff --git a/main.html b/main.html
index 7a61d6f..c3b9dfd 100644
--- a/main.html
+++ b/main.html
@@ -1349,13 +1349,13 @@
     </script>
 
     <!-- v4.5.63: A2A scratchpad fixes + agent-register arg fix -->
-<script src="modules/core/core-bundle.js?v=v4.6.46"></script>
+<script src="modules/core/core-bundle.js?v=v4.6.47"></script>
 
     <!-- Async dependency-ordered module loader now only initializes terminal and loads optional ext modules -->
-<script src="modules/core/module-loader.js?v=v4.6.46"></script>
+<script src="modules/core/module-loader.js?v=v4.6.47"></script>
 
     <!-- Critical ext bundle loads AFTER webx11 exists so registerModule calls work -->
-<script src="modules/ext/critical-ext-bundle.js?v=v4.6.46"></script>
+<script src="modules/ext/critical-ext-bundle.js?v=v4.6.47"></script>
 
 <script>
 // ═══ ONE TRUE VERSION SOURCE ═══
diff --git a/modules/core/core-bundle.js b/modules/core/core-bundle.js
index 4a2416c..c4bf64e 100644
--- a/modules/core/core-bundle.js
+++ b/modules/core/core-bundle.js
@@ -8218,10 +8218,10 @@ buildSystemPrompt() {
                     }
                 } catch (e) {}
                 base += '\n\n[File reads, tools, verification]\n' +
-                    '- read_file returns at most 8000 chars. If the reply says truncated=true, call read_file again with offset=<next_offset> before concluding anything about the file.\n' +
+                    '- read_file returns 8000 chars by default (limit up to 16000). If the reply says truncated=true, call read_file again with offset=<next_offset> before concluding anything about the file.\n' +
                     '- Prefer targeted reads over whole files: run_command with grep -n "<pattern>" <file>, sed -n \'10,40p\' <file>, head -40 <file>, tail -40 <file>, wc -l <file>.\n' +
                     '- sed without -i is read-only and allowed; sed -i, rm, mv and writes need user confirmation.\n' +
-                    '- After ANY write or edit: re-read the changed lines, then run a check (test, node scripts/lint-no-undef.py, or a curl) and say in one line what you verified.\n' +
+                    '- After ANY write or edit: re-read the changed lines (cat or sed -n) and run a check that exists in THIS terminal — `test` for the suite, `grep -n` to confirm the new line, `wc -l` for size. There is no shell, no node and no curl here.\n' +
                     '- If a tool fails twice, change approach (different tool, smaller scope) instead of retrying the same call.';
                 return base;
             }
@@ -9006,17 +9006,20 @@ buildSystemPrompt() {
                     }
                 } catch (e) { /* world model not available */ }
 
-                loopResult = await this.agenticLoop(messages, opts);
+                const _traceTag = opts && opts.agentTag ? opts.agentTag : 'main';
+                try {
+                    loopResult = await this.agenticLoop(messages, opts);
+                } catch (_loopErr) {
+                    this._flushToolTrace(sess, _traceTag);   // never leave stale notes behind
+                    throw _loopErr;
+                }
                 sess.context.push({ role: 'user', content: userMessage });
                 if (loopResult.finalAnswer) {
                     sess.context.push({ role: 'assistant', content: loopResult.finalAnswer });
                 }
                 // Durable trace of what was read/run this turn (2026-10-21) — without this,
                 // every later turn started blind and the model re-read the same files.
-                if (this._turnObservations && this._turnObservations.length) {
-                    sess.context.push({ role: 'assistant', content: '⚙️ tool trace: ' + this._turnObservations.join(' | ') });
-                }
-                this._turnObservations = [];
+                this._flushToolTrace(sess, _traceTag);
                 if (sess.context.length > this.maxContext * 2) {
                     sess.context.splice(0, sess.context.length - this.maxContext * 2);
                 }
@@ -9381,15 +9384,43 @@ buildSystemPrompt() {
             }
             // 2026-10-21: tool results lived only inside the current turn, so the next turn
             // had no idea what the model had already read or run. Keep a compact per-turn
-            // trace and persist it into the session context (see the turn-end push).
-            _noteObservation(name, args, result) {
+            // trace (per agent tag) and persist it into the session context.
+            // Review-hardened: no file bodies, no untrusted web output, no secrets.
+            _noteObservation(name, args, result, tag) {
                 try {
-                    if (!this._turnObservations) this._turnObservations = [];
-                    if (this._turnObservations.length >= 12) return;
-                    const argStr = (typeof args === 'object' && args !== null) ? JSON.stringify(args) : String(args || '');
-                    const note = (name + '(' + argStr + ') → ' + String(result === undefined || result === null ? '' : result))
-                        .replace(/\s+/g, ' ').trim();
-                    this._turnObservations.push(note.slice(0, 240));
+                    if (/^(web_search|web_fetch|scrape|fetch)$/.test(name)) return; // injection surface
+                    const key = tag || 'main';
+                    if (!this._turnObservations) this._turnObservations = {};
+                    const buf = this._turnObservations[key] || (this._turnObservations[key] = []);
+                    const clean = s => String(s === undefined || s === null ? '' : s)
+                        .replace(/\s+/g, ' ')
+                        .replace(/(token|key|secret|password|bearer)[^,}\s]*/ig, '$1:[redacted]');
+                    let argStr = (typeof args === 'object' && args !== null) ? JSON.stringify(args) : String(args || '');
+                    argStr = clean(argStr).slice(0, 80);
+                    let note;
+                    if (name === 'read_file') {
+                        let p = {};
+                        try { p = JSON.parse(String(result)) || {}; } catch (e) {}
+                        note = 'read_file ' + (p.path || argStr) + ' [' + (p.offset || 0) + '+' + (p.returned || 0) +
+                               '/' + (p.total_chars === undefined ? '?' : p.total_chars) +
+                               (p.truncated ? ', next=' + p.next_offset : '') + ']';
+                    } else {
+                        note = name + '(' + argStr + ') → ' + clean(String(result)).slice(0, 120);
+                    }
+                    buf.push(note.slice(0, 240));
+                    // Ring buffer: keep the first 2 (the goal) and the last 10 (verification).
+                    if (buf.length > 12) buf.splice(2, 1);
+                } catch (_e) {}
+            }
+            // Push this turn's trace into the session context and clear the buffer.
+            _flushToolTrace(sess, tag) {
+                try {
+                    const key = tag || 'main';
+                    const buf = (this._turnObservations && this._turnObservations[key]) || [];
+                    if (buf.length && sess && sess.context) {
+                        sess.context.push({ role: 'system', content: '⚙️ tool trace: ' + buf.join(' | ') });
+                    }
+                    if (this._turnObservations) delete this._turnObservations[key];
                 } catch (_e) {}
             }
             _escapeHtml(str) {
@@ -9608,9 +9639,14 @@ buildSystemPrompt() {
                         for (const tc of res.tool_calls) {
                             const sig = tc.function.name + ':' + (tc.function.arguments || '');
                             let entry = _seenCalls.get(sig);
-                            if (!entry || (now - entry.ts) > _loopWindow || entry.epoch !== this._toolEpoch) {
-                                // A mutation since the last attempt invalidates the block: re-run instead of replaying.
+                            if (!entry || (now - entry.ts) > _loopWindow) {
                                 entry = { count: 0, ts: now, resultHash: null, blocked: false, lastResult: null, epoch: this._toolEpoch };
+                            } else if (entry.epoch !== this._toolEpoch) {
+                                // A mutation since the last attempt unblocks the replay, but the
+                                // repeat counter must survive: resetting it let an identical
+                                // mutating call loop forever (review 2026-10-21).
+                                entry.epoch = this._toolEpoch;
+                                entry.blocked = false;
                             }
                             entry.count++;
                             entry.ts = now;
@@ -9739,7 +9775,7 @@ buildSystemPrompt() {
                                 }
                             } catch (_e) {}
                             messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
-                            this._noteObservation(fn.name, fn.arguments, toolResult);
+                            this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                         }
                         // Only block on destructive tools
                         if (destructiveTools.length > 0) {
@@ -9814,11 +9850,12 @@ buildSystemPrompt() {
                                         }
                                     } catch (_e) {}
                                     messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
-                                    this._noteObservation(fn.name, fn.arguments, toolResult);
+                                    this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                                 }
                             } else {
                                 messages.push({
                                     role: 'tool',
+                                    tool_call_id: (destructiveTools[0] && destructiveTools[0].id) || undefined,
                                     content: `Destructive actions (${destructiveTools.map(tc => tc.function.name).join(', ')}) denied by user. Non-destructive actions already executed.`
                                 });
                             }
@@ -10436,9 +10473,19 @@ buildSystemPrompt() {
                             // bigger file unreadable — the model could neither finish it nor
                             // resume. Now the caller pages with offset/limit and gets next_offset.
                             const total = content.length;
-                            const offset = Math.max(0, Math.min(parseInt(a.offset, 10) || 0, total));
-                            const limit = Math.max(1000, Math.min(parseInt(a.limit, 10) || 8000, 32000));
-                            const slice = content.slice(offset, offset + limit);
+                            let offset = Math.max(0, Math.min(parseInt(a.offset, 10) || 0, total));
+                            const limit = Math.max(500, Math.min(parseInt(a.limit, 10) || 8000, 16000));
+                            // Never split a surrogate pair (emoji/CJK-ext) across pages:
+                            // a lone surrogate makes JSON.stringify emit an invalid body.
+                            if (offset > 0) {
+                                const prev = content.charCodeAt(offset - 1);
+                                if (prev >= 0xD800 && prev <= 0xDBFF) offset += 1;
+                            }
+                            let slice = content.slice(offset, offset + limit);
+                            if (slice.length && slice.charCodeAt(slice.length - 1) >= 0xD800 &&
+                                slice.charCodeAt(slice.length - 1) <= 0xDBFF) {
+                                slice = slice.slice(0, -1);
+                            }
                             const nextOffset = offset + slice.length < total ? offset + slice.length : null;
                             return JSON.stringify({
                                 path: path, offset: offset, returned: slice.length,
@@ -12010,7 +12057,7 @@ buildSystemPrompt() {
                     (m.role === 'user' ? '👤' : m.role === 'assistant' ? '🤖' : '⚙️') + ': ' +
                     (m.content || '').substring(0, 400)
                 ).join(' | ');
-                let compressed = `[Earlier context (${oldMsgs.length} messages): ${summaryText}]`;
+                let compressed = `[Earlier context (${oldMsgs.length} messages): ${summaryText.slice(0, 4000)}]`;
                 // Tool traces must survive compression, so the AI one-sentence summary is only
                 // used when there are none to lose (2026-10-21).
                 const _traces = oldMsgs.filter(m => /^⚙️ tool trace:/.test(m.content || ''));
diff --git a/modules/core/module-aiassist.js b/modules/core/module-aiassist.js
index 90edf73..8d1fc9d 100644
--- a/modules/core/module-aiassist.js
+++ b/modules/core/module-aiassist.js
@@ -121,10 +121,10 @@ buildSystemPrompt() {
                     }
                 } catch (e) {}
                 base += '\n\n[File reads, tools, verification]\n' +
-                    '- read_file returns at most 8000 chars. If the reply says truncated=true, call read_file again with offset=<next_offset> before concluding anything about the file.\n' +
+                    '- read_file returns 8000 chars by default (limit up to 16000). If the reply says truncated=true, call read_file again with offset=<next_offset> before concluding anything about the file.\n' +
                     '- Prefer targeted reads over whole files: run_command with grep -n "<pattern>" <file>, sed -n \'10,40p\' <file>, head -40 <file>, tail -40 <file>, wc -l <file>.\n' +
                     '- sed without -i is read-only and allowed; sed -i, rm, mv and writes need user confirmation.\n' +
-                    '- After ANY write or edit: re-read the changed lines, then run a check (test, node scripts/lint-no-undef.py, or a curl) and say in one line what you verified.\n' +
+                    '- After ANY write or edit: re-read the changed lines (cat or sed -n) and run a check that exists in THIS terminal — `test` for the suite, `grep -n` to confirm the new line, `wc -l` for size. There is no shell, no node and no curl here.\n' +
                     '- If a tool fails twice, change approach (different tool, smaller scope) instead of retrying the same call.';
                 return base;
             }
@@ -909,17 +909,20 @@ buildSystemPrompt() {
                     }
                 } catch (e) { /* world model not available */ }
 
-                loopResult = await this.agenticLoop(messages, opts);
+                const _traceTag = opts && opts.agentTag ? opts.agentTag : 'main';
+                try {
+                    loopResult = await this.agenticLoop(messages, opts);
+                } catch (_loopErr) {
+                    this._flushToolTrace(sess, _traceTag);   // never leave stale notes behind
+                    throw _loopErr;
+                }
                 sess.context.push({ role: 'user', content: userMessage });
                 if (loopResult.finalAnswer) {
                     sess.context.push({ role: 'assistant', content: loopResult.finalAnswer });
                 }
                 // Durable trace of what was read/run this turn (2026-10-21) — without this,
                 // every later turn started blind and the model re-read the same files.
-                if (this._turnObservations && this._turnObservations.length) {
-                    sess.context.push({ role: 'assistant', content: '⚙️ tool trace: ' + this._turnObservations.join(' | ') });
-                }
-                this._turnObservations = [];
+                this._flushToolTrace(sess, _traceTag);
                 if (sess.context.length > this.maxContext * 2) {
                     sess.context.splice(0, sess.context.length - this.maxContext * 2);
                 }
@@ -1284,15 +1287,43 @@ buildSystemPrompt() {
             }
             // 2026-10-21: tool results lived only inside the current turn, so the next turn
             // had no idea what the model had already read or run. Keep a compact per-turn
-            // trace and persist it into the session context (see the turn-end push).
-            _noteObservation(name, args, result) {
+            // trace (per agent tag) and persist it into the session context.
+            // Review-hardened: no file bodies, no untrusted web output, no secrets.
+            _noteObservation(name, args, result, tag) {
                 try {
-                    if (!this._turnObservations) this._turnObservations = [];
-                    if (this._turnObservations.length >= 12) return;
-                    const argStr = (typeof args === 'object' && args !== null) ? JSON.stringify(args) : String(args || '');
-                    const note = (name + '(' + argStr + ') → ' + String(result === undefined || result === null ? '' : result))
-                        .replace(/\s+/g, ' ').trim();
-                    this._turnObservations.push(note.slice(0, 240));
+                    if (/^(web_search|web_fetch|scrape|fetch)$/.test(name)) return; // injection surface
+                    const key = tag || 'main';
+                    if (!this._turnObservations) this._turnObservations = {};
+                    const buf = this._turnObservations[key] || (this._turnObservations[key] = []);
+                    const clean = s => String(s === undefined || s === null ? '' : s)
+                        .replace(/\s+/g, ' ')
+                        .replace(/(token|key|secret|password|bearer)[^,}\s]*/ig, '$1:[redacted]');
+                    let argStr = (typeof args === 'object' && args !== null) ? JSON.stringify(args) : String(args || '');
+                    argStr = clean(argStr).slice(0, 80);
+                    let note;
+                    if (name === 'read_file') {
+                        let p = {};
+                        try { p = JSON.parse(String(result)) || {}; } catch (e) {}
+                        note = 'read_file ' + (p.path || argStr) + ' [' + (p.offset || 0) + '+' + (p.returned || 0) +
+                               '/' + (p.total_chars === undefined ? '?' : p.total_chars) +
+                               (p.truncated ? ', next=' + p.next_offset : '') + ']';
+                    } else {
+                        note = name + '(' + argStr + ') → ' + clean(String(result)).slice(0, 120);
+                    }
+                    buf.push(note.slice(0, 240));
+                    // Ring buffer: keep the first 2 (the goal) and the last 10 (verification).
+                    if (buf.length > 12) buf.splice(2, 1);
+                } catch (_e) {}
+            }
+            // Push this turn's trace into the session context and clear the buffer.
+            _flushToolTrace(sess, tag) {
+                try {
+                    const key = tag || 'main';
+                    const buf = (this._turnObservations && this._turnObservations[key]) || [];
+                    if (buf.length && sess && sess.context) {
+                        sess.context.push({ role: 'system', content: '⚙️ tool trace: ' + buf.join(' | ') });
+                    }
+                    if (this._turnObservations) delete this._turnObservations[key];
                 } catch (_e) {}
             }
             _escapeHtml(str) {
@@ -1511,9 +1542,14 @@ buildSystemPrompt() {
                         for (const tc of res.tool_calls) {
                             const sig = tc.function.name + ':' + (tc.function.arguments || '');
                             let entry = _seenCalls.get(sig);
-                            if (!entry || (now - entry.ts) > _loopWindow || entry.epoch !== this._toolEpoch) {
-                                // A mutation since the last attempt invalidates the block: re-run instead of replaying.
+                            if (!entry || (now - entry.ts) > _loopWindow) {
                                 entry = { count: 0, ts: now, resultHash: null, blocked: false, lastResult: null, epoch: this._toolEpoch };
+                            } else if (entry.epoch !== this._toolEpoch) {
+                                // A mutation since the last attempt unblocks the replay, but the
+                                // repeat counter must survive: resetting it let an identical
+                                // mutating call loop forever (review 2026-10-21).
+                                entry.epoch = this._toolEpoch;
+                                entry.blocked = false;
                             }
                             entry.count++;
                             entry.ts = now;
@@ -1642,7 +1678,7 @@ buildSystemPrompt() {
                                 }
                             } catch (_e) {}
                             messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
-                            this._noteObservation(fn.name, fn.arguments, toolResult);
+                            this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                         }
                         // Only block on destructive tools
                         if (destructiveTools.length > 0) {
@@ -1717,11 +1753,12 @@ buildSystemPrompt() {
                                         }
                                     } catch (_e) {}
                                     messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
-                                    this._noteObservation(fn.name, fn.arguments, toolResult);
+                                    this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                                 }
                             } else {
                                 messages.push({
                                     role: 'tool',
+                                    tool_call_id: (destructiveTools[0] && destructiveTools[0].id) || undefined,
                                     content: `Destructive actions (${destructiveTools.map(tc => tc.function.name).join(', ')}) denied by user. Non-destructive actions already executed.`
                                 });
                             }
@@ -2339,9 +2376,19 @@ buildSystemPrompt() {
                             // bigger file unreadable — the model could neither finish it nor
                             // resume. Now the caller pages with offset/limit and gets next_offset.
                             const total = content.length;
-                            const offset = Math.max(0, Math.min(parseInt(a.offset, 10) || 0, total));
-                            const limit = Math.max(1000, Math.min(parseInt(a.limit, 10) || 8000, 32000));
-                            const slice = content.slice(offset, offset + limit);
+                            let offset = Math.max(0, Math.min(parseInt(a.offset, 10) || 0, total));
+                            const limit = Math.max(500, Math.min(parseInt(a.limit, 10) || 8000, 16000));
+                            // Never split a surrogate pair (emoji/CJK-ext) across pages:
+                            // a lone surrogate makes JSON.stringify emit an invalid body.
+                            if (offset > 0) {
+                                const prev = content.charCodeAt(offset - 1);
+                                if (prev >= 0xD800 && prev <= 0xDBFF) offset += 1;
+                            }
+                            let slice = content.slice(offset, offset + limit);
+                            if (slice.length && slice.charCodeAt(slice.length - 1) >= 0xD800 &&
+                                slice.charCodeAt(slice.length - 1) <= 0xDBFF) {
+                                slice = slice.slice(0, -1);
+                            }
                             const nextOffset = offset + slice.length < total ? offset + slice.length : null;
                             return JSON.stringify({
                                 path: path, offset: offset, returned: slice.length,
@@ -3913,7 +3960,7 @@ buildSystemPrompt() {
                     (m.role === 'user' ? '👤' : m.role === 'assistant' ? '🤖' : '⚙️') + ': ' +
                     (m.content || '').substring(0, 400)
                 ).join(' | ');
-                let compressed = `[Earlier context (${oldMsgs.length} messages): ${summaryText}]`;
+                let compressed = `[Earlier context (${oldMsgs.length} messages): ${summaryText.slice(0, 4000)}]`;
                 // Tool traces must survive compression, so the AI one-sentence summary is only
                 // used when there are none to lose (2026-10-21).
                 const _traces = oldMsgs.filter(m => /^⚙️ tool trace:/.test(m.content || ''));
diff --git a/modules/core/module-loader.js b/modules/core/module-loader.js
index 6c3b48e..ccd92d8 100644
--- a/modules/core/module-loader.js
+++ b/modules/core/module-loader.js
@@ -4,7 +4,7 @@
     // Core module bundle loaded synchronously in main.html.
     // This loader only handles version cache-busting, splash, terminal init, and optional ext modules.
 
-    const LATEST_VER = 'v4.6.46';
+    const LATEST_VER = 'v4.6.47';
     const CACHE = '?v=' + LATEST_VER;
     const stored = localStorage.getItem('webx11_cache_ver');
     if (stored !== LATEST_VER) {
-- 
cgit v1.2.3