fix(newos4): bound one turn's tool context, honest skill-create, quarantine a corrupt memory graph

The three items the last review and probes left open, each with a test that is red on HEAD.

- MAINT-023 (module-aiassist.js): one agentic turn grew without limit — every role:'tool' body
  was pushed verbatim and only this.context was trimmed BETWEEN turns (measured 6 -> 112 705
  tokens across 8 model calls against maxContextTokens 12 000). Observations are now clipped to
  maxToolResultChars (6 000) with a note saying how much was withheld and to re-read with
  offset/limit, and _boundToolContext() elides the oldest tool bodies once the running total
  passes toolContextBudgetChars (24 000), always keeping the newest 4 verbatim. The ceiling is
  turn-independent: the same ~26 k chars at 8 turns and at 20.
- MAINT-024 (modules/ext/skill-pack-core.js): the command printed "✅ Skill created" with
  fs.writeFile()'s boolean discarded (probe: fresh VFS, success printed, no file), stripped
  invalid characters instead of rejecting the name (My-Skill -> y-kill, DevOps_Thing ->
  evpshing), and treated an unreadable /path argument as the skill body. Now: kebab-case is
  validated (uppercase normalises and the final name is reported; anything else is refused),
  /skills is created on first use, the write and index results are checked, and an unreadable
  body path is an error.
- MAINT-003 (module-memory-graph.js): a corrupt store looked exactly like a fresh install —
  fsReadJSON returns null on a parse failure, so 85 nodes could vanish with no warning and no
  copy. The probe confirmed the loss: seed 6 nodes, truncate nodes.json and export.json, reload
  -> 1 node, no message, nothing in .trash. _quarantineCorrupt() now keeps the unreadable file
  as <name>.corrupt, warns, and then loads from whatever remains.

Tests: test-tool-context-bound.js (11), test-skill-create.js (16), test-memory-graph-recovery.js
(8). RED on HEAD: tool-context crashes on _clipToolResult (exit 1), skill-create 16 assertions
fail on the missing file/index, memory-graph B1/B2/C1 fail (no warning, no .corrupt copy).
Suite now 13/13 files green; maintain --fix reports "node tests 13/13 pass".
Live on v4.6.107: skill-create My-Skill -> ✅ "my-skill"; DevOps_Thing -> Invalid skill name;
/does/not/exist.md -> Cannot read skill body; only my-skill in the index.

1 files changed, 67 insertions, 2 deletions

diff --git a/modules/core/core-bundle.js b/modules/core/core-bundle.js
index a057332..c06af5f 100644
--- a/modules/core/core-bundle.js
+++ b/modules/core/core-bundle.js
@@ -9837,6 +9837,41 @@ buildSystemPrompt() {
                 }
                 return result.substring(0, maxLen) + '...(truncated from ' + result.length + ' chars)';
             }
+            // MAINT-023: one agentic turn used to grow without limit — every role:'tool' body was
+            // pushed verbatim and only this.context was trimmed BETWEEN turns, so a measured turn
+            // went 6 → 112 705 tokens across 8 model calls (maxContextTokens is 12 000). Two
+            // limits keep a single turn inside the window: clip each observation as it arrives,
+            // and elide the oldest ones once the running total passes the budget.
+            _clipToolResult(result) {
+                const text = typeof result === 'string' ? result : JSON.stringify(result);
+                const max = this.maxToolResultChars || 6000;
+                if (!text || text.length <= max) return text;
+                return text.slice(0, max)
+                    + '\n…[clipped: ' + (text.length - max) + ' of ' + text.length
+                    + ' chars withheld — re-read with offset/limit if you need the rest]';
+            }
+            _boundToolContext(messages) {
+                const keepVerbatim = this.toolResultsVerbatim || 4;
+                const budget = this.toolContextBudgetChars || 24000;
+                const idx = [];
+                for (let i = 0; i < messages.length; i++) {
+                    if (messages[i] && messages[i].role === 'tool') idx.push(i);
+                }
+                if (!idx.length) return messages;
+                const recent = new Set(idx.slice(-keepVerbatim));
+                let total = 0;
+                for (const i of idx) total += String(messages[i].content || '').length;
+                // Oldest first, never the recent few: those are what the model is working on.
+                for (const i of idx) {
+                    if (total <= budget) break;
+                    if (recent.has(i)) continue;
+                    const len = String(messages[i].content || '').length;
+                    const stub = '[earlier tool result elided — ' + len + ' chars; call the tool again if you still need it]';
+                    messages[i].content = stub;
+                    total -= len - stub.length;
+                }
+                return messages;
+            }
             // 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 (per agent tag) and persist it into the session context.
@@ -10067,6 +10102,8 @@ buildSystemPrompt() {
                         messages.push({ role: 'system', content: finalPrompt });
                     }
                     const thinkStart = Date.now();
+                    // MAINT-023: keep the in-flight tool context bounded before every model call.
+                    this._boundToolContext(messages);
                     const res = await this.terminal.ollamaManager.chatWithTools(messages, this.getToolDefinitions());
                     const thinkTime = Math.round((Date.now() - thinkStart) / 1000);
                     // Store raw response for debug-ai-last
@@ -10236,7 +10273,9 @@ buildSystemPrompt() {
                                     _eR.lastResult = typeof toolResult === 'string' ? toolResult : JSON.stringify(toolResult);
                                 }
                             } catch (_e) {}
-                            messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
+                            // MAINT-023: clip on the way in — a single read_file page is 16 000
+                            // chars and several of them used to enter the context verbatim.
+                            messages.push({ role: 'tool', tool_call_id: tc.id, content: this._clipToolResult(toolResult) });
                             this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                         }
                         // Only block on destructive tools
@@ -10311,7 +10350,8 @@ buildSystemPrompt() {
                                             _eD.lastResult = typeof toolResult === 'string' ? toolResult : JSON.stringify(toolResult);
                                         }
                                     } catch (_e) {}
-                                    messages.push({ role: 'tool', tool_call_id: tc.id, content: toolResult });
+                                    // MAINT-023: clip on the way in (same rule as the main loop).
+                                    messages.push({ role: 'tool', tool_call_id: tc.id, content: this._clipToolResult(toolResult) });
                                     this._noteObservation(fn.name, fn.arguments, toolResult, _agentTag);
                                 }
                             } else {
@@ -18163,6 +18203,12 @@ class MemoryGraphModule {
         try {
             const fs = this.terminal.fs;
             if (!fs) return;
+            // MAINT-003: a corrupt store used to look exactly like a fresh one — fsReadJSON
+            // returns null on a parse failure, so the graph started empty with no warning and no
+            // copy of what was there. Quarantine anything unreadable first, then load what's left.
+            this._quarantineCorrupt('export.json');
+            this._quarantineCorrupt('nodes.json');
+            this._quarantineCorrupt('edges.json');
 
             // Try loading from export first
             const exportData = fs.fsReadJSON(`${this._indexPath}export.json`);
@@ -18215,6 +18261,25 @@ class MemoryGraphModule {
         }
     }
 
+    // MAINT-003: keep an unreadable store aside instead of starting fresh over it.
+    _quarantineCorrupt(name) {
+        try {
+            const fs = this.terminal.fs;
+            const p = `${this._indexPath}${name}`;
+            const raw = fs.readFile(p);
+            if (typeof raw !== 'string' || !raw.length) return false;
+            try { JSON.parse(raw); return false; } catch (e) {
+                const kept = p + '.corrupt';
+                fs.writeFile(kept, raw);
+                this.terminal.print(`⚠️ Memory graph store ${name} is unreadable — kept a copy at ${kept} and starting from what remains`, 'warning');
+                if (typeof fs.deleteFile === 'function') fs.deleteFile(p);
+                return true;
+            }
+        } catch (e) {
+            return false;
+        }
+    }
+
     _saveNow() {
         try {
             const fs = this.terminal.fs;