clementine-agent 1.18.132 → 1.18.134

package/dist/agent/{self-improve-loop.d.ts → failure-fix-consumer.d.ts}

@@ -98,7 +98,7 @@ export interface SelfImproveLoopOptions {
     allowAutoApplySafeFixes?: boolean;
 }
 export declare function classifyFailure(recentErrors: string[]): FixRecipe;
-export declare class SelfImproveLoop {
+export declare class FailureFixConsumer {
     private readonly tickMs;
     private readonly triggersDir;
     private readonly pendingDir;
@@ -133,4 +133,4 @@ export declare class SelfImproveLoop {
     private processOne;
     private notifyAgent;
 }
-//# sourceMappingURL=self-improve-loop.d.ts.map
+//# sourceMappingURL=failure-fix-consumer.d.ts.map

package/dist/agent/{self-improve-loop.js → failure-fix-consumer.js}

@@ -29,7 +29,7 @@ import path from 'node:path';
 import matter from 'gray-matter';
 import pino from 'pino';
 import { AGENTS_DIR, BASE_DIR, SYSTEM_DIR } from '../config.js';
-const logger = pino({ name: 'clementine.self-improve-loop' });
+const logger = pino({ name: 'clementine.failure-fix-consumer' });
 /**
  * Fallback tick interval. The loop is primarily event-driven via fs.watch
  * on the triggers directory — this is just a slow safety net for cases
@@ -231,7 +231,7 @@ function writePendingChange(record, dir) {
     return file;
 }
 // ── Main loop ────────────────────────────────────────────────────────
-export class SelfImproveLoop {
+export class FailureFixConsumer {
     tickMs;
     triggersDir;
     pendingDir;
@@ -503,4 +503,4 @@ export class SelfImproveLoop {
         }
     }
 }
-//# sourceMappingURL=self-improve-loop.js.map
+//# sourceMappingURL=failure-fix-consumer.js.map

package/dist/agent/self-improve.js

@@ -411,12 +411,18 @@ export class SelfImproveLoop {
                         consecutiveLow++;
                         continue;
                     }
-                    // Diversity safety net: skip if hypothesis targets an over-represented area:target
+                    // Diversity safety net: skip if hypothesis targets an over-represented area:target.
+                    // 1.18.134 — loosened cap from 3 to 5. The old cap caused the
+                    // loop to plateau immediately whenever it had ~3 ideas about
+                    // SOUL.md (which is a frequent attractor). At 5 the
+                    // hypothesizer gets a few more swings at the same area before
+                    // diversity kicks in — but still avoids monomania.
+                    const DIVERSITY_CAP = 5;
                     const proposalKey = `${proposal.area}:${proposal.target}`;
                     const proposalCount = history.filter(e => `${e.area}:${e.target}` === proposalKey).length
                         + this.getPendingChanges().filter(p => `${p.area}:${p.target}` === proposalKey).length;
-                    if (proposalCount >= 3) {
-                        logger.warn({ area: proposal.area, target: proposal.target, count: proposalCount }, 'Hypothesis over-targeted — skipping');
+                    if (proposalCount >= DIVERSITY_CAP) {
+                        logger.warn({ area: proposal.area, target: proposal.target, count: proposalCount, cap: DIVERSITY_CAP }, 'Hypothesis over-targeted — skipping');
                         consecutiveLow++;
                         continue;
                     }
@@ -440,8 +446,37 @@ export class SelfImproveLoop {
                     const before = await this.readCurrentState(proposal.area, proposal.target);
                     // Step 5: Evaluate
                     const evaluation = await this.withTimeout(this.evaluate(before, proposal.proposedChange, proposal.hypothesis), 60_000);
-                    const score = evaluation?.score ?? 0;
-                    const normalizedScore = score / 10; // Convert 0-10 to 0-1
+                    const llmScore = evaluation?.score ?? 0;
+                    const normalizedLlmScore = llmScore / 10; // Convert 0-10 to 0-1
+                    // 1.18.134 — blend the LLM evaluator score with an objective
+                    // signal pulled from real metrics. Karpathy's autoresearch uses
+                    // ONE objective metric (val_bpb); Clementine's LLM-only score
+                    // can drift, especially when proposals affect SOUL.md or
+                    // cosmetic prompt fields. The objective floor is computed from
+                    // the baseline metrics gathered at the start of THIS run:
+                    //
+                    //   objective = cronSuccessRate * feedbackPositiveRatio
+                    //
+                    // Both 0..1; product preserves the multiplicative penalty of
+                    // either metric being weak. We weight 70% LLM / 30% objective
+                    // — the LLM still drives most decisions, but a proposal can't
+                    // sail through on a 9/10 evaluator score when feedback is at
+                    // 35% positive (which is the current reality). Set the env
+                    // var SELF_IMPROVE_OBJECTIVE_WEIGHT to override (0..1).
+                    const objectiveWeightRaw = parseFloat(process.env.SELF_IMPROVE_OBJECTIVE_WEIGHT ?? '0.3');
+                    const objectiveWeight = Number.isFinite(objectiveWeightRaw) && objectiveWeightRaw >= 0 && objectiveWeightRaw <= 1
+                        ? objectiveWeightRaw : 0.3;
+                    const llmWeight = 1 - objectiveWeight;
+                    const objectiveScore = (state.baselineMetrics.cronSuccessRate || 0)
+                        * (state.baselineMetrics.feedbackPositiveRatio || 0);
+                    const normalizedScore = (normalizedLlmScore * llmWeight) + (objectiveScore * objectiveWeight);
+                    const score = normalizedScore * 10; // For display + reason text
+                    logger.debug({
+                        llm: normalizedLlmScore.toFixed(3),
+                        objective: objectiveScore.toFixed(3),
+                        blended: normalizedScore.toFixed(3),
+                        weight: objectiveWeight,
+                    }, 'Score blend');
                     const accepted = normalizedScore >= this.config.acceptThreshold;
                     // Surface gate: even when accepted, only score >= surfaceThreshold
                     // reaches the user's pending-changes inbox. Below that floor we

package/dist/cli/dashboard.js

@@ -7047,6 +7047,57 @@ If the tool returns nothing or errors, return an empty array \`[]\`.`,
             res.status(500).json({ error: String(err) });
         }
     });
+    // 1.18.132 — list top-level files in a registered project. Used by
+    // the Skill Builder's Files sidebar tab to surface clickable file
+    // paths that the user can insert into their skill body.
+    // Hardened: the path query param MUST match a registered project's
+    // path exactly (no path traversal, no scanning arbitrary directories).
+    app.get('/api/projects/files', (req, res) => {
+        try {
+            const projPath = String(req.query.path ?? '');
+            if (!projPath)
+                return res.status(400).json({ ok: false, error: 'path query param required' });
+            const projects = loadProjectsMeta();
+            const found = projects.find((p) => p.path === projPath);
+            if (!found)
+                return res.status(404).json({ ok: false, error: 'path is not a registered project' });
+            if (!existsSync(projPath))
+                return res.json({ ok: true, files: [], note: 'project path does not exist on disk' });
+            let entries;
+            try {
+                entries = readdirSync(projPath);
+            }
+            catch (err) {
+                return res.status(500).json({ ok: false, error: 'failed to list directory: ' + String(err) });
+            }
+            // Skip hidden + node_modules + standard noise. Up to 50 entries
+            // (the sidebar caps at 50 anyway). One level deep — the panel is
+            // an entry point; users can dig further from their editor.
+            const NOISE = new Set(['node_modules', '.git', '.DS_Store', 'dist', 'build', '.next', '.cache']);
+            const out = [];
+            for (const entry of entries.sort()) {
+                if (entry.startsWith('.'))
+                    continue;
+                if (NOISE.has(entry))
+                    continue;
+                const abs = path.join(projPath, entry);
+                let st;
+                try {
+                    st = statSync(abs);
+                }
+                catch {
+                    continue;
+                }
+                out.push({ relPath: entry, isDir: st.isDirectory(), sizeBytes: st.isFile() ? st.size : 0 });
+                if (out.length >= 50)
+                    break;
+            }
+            res.json({ ok: true, files: out });
+        }
+        catch (err) {
+            res.status(500).json({ ok: false, error: String(err) });
+        }
+    });
     // ── Available Tools ──────────────────────────────────────────
     app.get('/api/available-tools', async (_req, res) => {
         try {
@@ -28178,6 +28229,7 @@ async function openSkillBuilder(skillName) {
       +     '<span id="sb-save-status" style="font-size:11px;color:var(--text-muted)"></span>'
       +   '</div>'
       +   '<div style="display:flex;align-items:center;gap:6px">'
+      +     '<button onclick="sbRunSkillTest()" id="sb-test-btn" style="font-size:12px;padding:7px 12px;border:1px solid var(--green);border-radius:6px;background:transparent;color:var(--green);cursor:pointer;font-weight:500" title="Fire this skill once and stream the result inline (toast on completion)">▶ Test run</button>'
       +     '<button onclick="sbSaveCurrent()" id="sb-save-btn" class="btn-primary" style="font-size:12px;padding:7px 14px;border:none;border-radius:6px;background:var(--accent);color:#fff;font-weight:500;cursor:pointer" disabled>Save (⌘S)</button>'
       +     '<button onclick="closeSkillBuilder()" style="font-size:12px;padding:7px 12px;border:1px solid var(--border);border-radius:6px;background:transparent;color:var(--text-primary);cursor:pointer">Close</button>'
       +   '</div>'
@@ -28203,6 +28255,8 @@ async function openSkillBuilder(skillName) {
       +   '<div style="border-left:1px solid var(--border);background:var(--bg-secondary);display:flex;flex-direction:column;min-height:0">'
       +     '<div style="padding:8px 6px 0;display:flex;gap:2px;border-bottom:1px solid var(--border)">'
       +       '<button onclick="sbSwitchTab(\\x27tools\\x27)" id="sb-tab-tools" class="sb-tab" style="flex:1;font-size:11px;padding:8px 4px;border:none;background:transparent;color:var(--text-muted);cursor:pointer;border-bottom:2px solid transparent;font-weight:500">Tools</button>'
+      +       '<button onclick="sbSwitchTab(\\x27files\\x27)" id="sb-tab-files" class="sb-tab" style="flex:1;font-size:11px;padding:8px 4px;border:none;background:transparent;color:var(--text-muted);cursor:pointer;border-bottom:2px solid transparent;font-weight:500">Files</button>'
+      +       '<button onclick="sbSwitchTab(\\x27memory\\x27)" id="sb-tab-memory" class="sb-tab" style="flex:1;font-size:11px;padding:8px 4px;border:none;background:transparent;color:var(--text-muted);cursor:pointer;border-bottom:2px solid transparent;font-weight:500">Memory</button>'
       +       '<button onclick="sbSwitchTab(\\x27skills\\x27)" id="sb-tab-skills" class="sb-tab" style="flex:1;font-size:11px;padding:8px 4px;border:none;background:transparent;color:var(--text-muted);cursor:pointer;border-bottom:2px solid transparent;font-weight:500">Skills</button>'
       +     '</div>'
       +     '<div style="padding:6px 10px;border-bottom:1px solid var(--border)">'
@@ -28389,10 +28443,10 @@ async function sbDeleteFile(relPath) {
   }
 }
 
-// ── Sidebar tabs (Tools / Skills) ────────────────────────────────────
+// ── Sidebar tabs (Tools / Files / Memory / Skills) ──────────────────
 function sbSwitchTab(tab) {
   window._sbActiveTab = tab;
-  ['tools', 'skills'].forEach(function(t) {
+  ['tools', 'files', 'memory', 'skills'].forEach(function(t) {
     var el = document.getElementById('sb-tab-' + t);
     if (el) {
       el.style.color = (t === tab) ? 'var(--accent)' : 'var(--text-muted)';
@@ -28408,6 +28462,10 @@ async function sbRenderSidebar() {
   var q = (document.getElementById('sb-sidebar-search')?.value || '').toLowerCase().trim();
   if (window._sbActiveTab === 'tools') {
     await sbRenderToolsTab(listEl, q);
+  } else if (window._sbActiveTab === 'files') {
+    await sbRenderFilesTab(listEl, q);
+  } else if (window._sbActiveTab === 'memory') {
+    await sbRenderMemoryTab(listEl, q);
   } else if (window._sbActiveTab === 'skills') {
     await sbRenderSkillsTab(listEl, q);
   }
@@ -28505,6 +28563,163 @@ async function sbRenderSkillsTab(listEl, q) {
   listEl.innerHTML = html;
 }
 
+// 1.18.132 — Files tab (Phase 2.5). Browses the user's projects.
+// Click a project to expand its file tree; click a file to insert
+// its absolute path at the cursor (so the skill body can reference
+// it via Read or Bash). Scoped to projectsData (already populated by
+// /api/projects on page load); no separate filesystem walk endpoint
+// needed in this iteration.
+async function sbRenderFilesTab(listEl, q) {
+  var projects = (typeof projectsData !== 'undefined' && Array.isArray(projectsData)) ? projectsData : [];
+  if (projects.length === 0) {
+    try {
+      var pr = await window.apiFetch('/api/projects');
+      var pd = await pr.json();
+      if (pd && Array.isArray(pd.projects)) projects = pd.projects;
+    } catch (_) { /* fall through to empty */ }
+  }
+  if (q) {
+    projects = projects.filter(function(p) {
+      return ((p.name || '') + ' ' + (p.path || '')).toLowerCase().indexOf(q) > -1;
+    });
+  }
+  if (projects.length === 0) {
+    listEl.innerHTML = '<div style="padding:18px;color:var(--text-muted);font-size:11px;text-align:center">' + (q ? 'No matches.' : 'No projects yet. Add one in Settings → Projects.') + '</div>';
+    return;
+  }
+  var html = '<div style="padding:6px 12px;font-size:10px;font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.04em;background:var(--bg-tertiary)">Projects (' + projects.length + ')</div>';
+  for (var i = 0; i < projects.length; i++) {
+    var p = projects[i];
+    // Click on row → insert the project path. Click on icon to load
+    // a child file list inline (lazy expand).
+    html += '<div style="border-bottom:1px solid var(--border-light)">'
+      + '<div onclick="sbInsertAtCursor(\\x27' + jsStr(p.path) + '\\x27)" style="padding:7px 12px;cursor:pointer;font-size:12px;display:flex;align-items:center;gap:6px" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27" title="Insert this absolute path at the cursor">'
+      +   '<span>📁</span>'
+      +   '<span style="font-weight:500;color:var(--text-primary);flex:1;overflow:hidden;text-overflow:ellipsis;white-space:nowrap">' + esc(p.name || p.path.split(\'/\').pop()) + '</span>'
+      +   '<button onclick="event.stopPropagation();sbToggleProjectFiles(\\x27' + jsStr(p.path) + '\\x27, this)" title="Browse files inside this project" style="background:none;border:1px solid var(--border);color:var(--text-muted);font-size:10px;padding:1px 6px;border-radius:3px;cursor:pointer">▸</button>'
+      + '</div>'
+      + '<div id="sb-files-children-' + i + '" data-project-path="' + esc(p.path) + '" style="display:none;font-size:11px;background:var(--bg-secondary);padding:4px 0"></div>'
+      + '</div>';
+  }
+  listEl.innerHTML = html;
+}
+
+// Lazy-load the child file list for a project. Hits a tiny endpoint
+// that returns the top-level files (max 50, no recursion) so we don't
+// blow the panel on huge projects.
+async function sbToggleProjectFiles(projectPath, btn) {
+  // Find the children container next to this button
+  var children = btn.parentElement.parentElement.querySelector('[data-project-path]');
+  if (!children) return;
+  if (children.style.display !== 'none') {
+    children.style.display = 'none';
+    btn.textContent = '▸';
+    return;
+  }
+  children.style.display = '';
+  btn.textContent = '▾';
+  if (!children.dataset.loaded) {
+    children.innerHTML = '<div style="padding:8px 24px;color:var(--text-muted);font-size:10px">Loading…</div>';
+    try {
+      var r = await window.apiFetch('/api/projects/files?path=' + encodeURIComponent(projectPath));
+      var d = await r.json();
+      if (!r.ok || !Array.isArray(d.files)) {
+        children.innerHTML = '<div style="padding:8px 24px;color:var(--text-muted);font-size:10px">No files surfaced (or endpoint unavailable).</div>';
+        return;
+      }
+      var html = '';
+      for (var i = 0; i < Math.min(d.files.length, 50); i++) {
+        var f = d.files[i];
+        var icon = f.isDir ? '📁' : '📄';
+        var fullPath = projectPath + '/' + f.relPath;
+        html += '<div onclick="sbInsertAtCursor(\\x27' + jsStr(fullPath) + '\\x27)" style="padding:4px 24px;cursor:pointer;display:flex;align-items:center;gap:6px" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27">'
+          + '<span>' + icon + '</span>'
+          + '<span style="flex:1;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;color:var(--text-secondary)">' + esc(f.relPath) + '</span>'
+          + '</div>';
+      }
+      if (d.files.length > 50) {
+        html += '<div style="padding:4px 24px;color:var(--text-muted);font-size:10px;font-style:italic">+ ' + (d.files.length - 50) + ' more (open the project in your editor for the rest)</div>';
+      }
+      children.innerHTML = html;
+      children.dataset.loaded = '1';
+    } catch (err) {
+      children.innerHTML = '<div style="padding:8px 24px;color:var(--red);font-size:10px">' + esc(String(err)) + '</div>';
+    }
+  }
+}
+
+// 1.18.132 — Memory tab (Phase 2.5). Surfaces three things the agent
+// can reach when the skill runs:
+//   1. Recent extractions (last facts the auto-extractor saved)
+//   2. Top-level MEMORY.md sections (h2 headers)
+//   3. A button to insert a memory_search call template
+// Click any item → inserts a contextual reference at the cursor so
+// the skill body can document or trigger a recall.
+async function sbRenderMemoryTab(listEl, q) {
+  var html = '';
+  // 1. Recent extractions
+  try {
+    var r = await window.apiFetch('/api/memory/writes/recent?limit=12');
+    var d = await r.json();
+    var writes = (d && Array.isArray(d.writes)) ? d.writes : (d && Array.isArray(d.entries) ? d.entries : []);
+    if (q) {
+      writes = writes.filter(function(w) {
+        return JSON.stringify(w).toLowerCase().indexOf(q) > -1;
+      });
+    }
+    if (writes.length > 0) {
+      html += '<div style="padding:6px 12px;font-size:10px;font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.04em;background:var(--bg-tertiary)">Recent extractions</div>';
+      for (var i = 0; i < Math.min(writes.length, 8); i++) {
+        var w = writes[i];
+        var preview = (w.content_preview || w.content || w.preview || w.text || '').slice(0, 100);
+        if (!preview) continue;
+        var insertion = 'Reference recent fact: "' + preview.replace(/"/g, '\\\\\\\\"').slice(0, 80) + '"';
+        html += '<div onclick="sbInsertAtCursor(\\x27' + jsStr(insertion) + '\\x27)" style="padding:7px 12px;cursor:pointer;font-size:12px;border-bottom:1px solid var(--border-light)" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27">'
+          + '<div style="color:var(--text-secondary);line-height:1.4;font-size:11px">' + esc(preview) + (preview.length >= 100 ? '…' : '') + '</div>'
+          + '</div>';
+      }
+    }
+  } catch (_) { /* memory writes endpoint may not be live in early daemons */ }
+  // 2. MEMORY.md slot headers
+  try {
+    var r2 = await window.apiFetch('/api/memory/md');
+    var d2 = await r2.json();
+    var content = (d2 && d2.content) || '';
+    var lines = content.split('\\n');
+    var sections = [];
+    for (var li = 0; li < lines.length; li++) {
+      var match = lines[li].match(/^##\\s+(.+)$/);
+      if (match) sections.push(match[1].trim());
+    }
+    if (q) {
+      sections = sections.filter(function(s) { return s.toLowerCase().indexOf(q) > -1; });
+    }
+    if (sections.length > 0) {
+      html += '<div style="padding:6px 12px;font-size:10px;font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.04em;background:var(--bg-tertiary);margin-top:4px">MEMORY.md sections</div>';
+      for (var si = 0; si < sections.length; si++) {
+        var s = sections[si];
+        var ins = 'Refer to MEMORY.md > "' + s + '" section.';
+        html += '<div onclick="sbInsertAtCursor(\\x27' + jsStr(ins) + '\\x27)" style="padding:6px 12px;cursor:pointer;font-size:12px;display:flex;align-items:center;gap:6px;border-bottom:1px solid var(--border-light)" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27">'
+          + '<span>📄</span>'
+          + '<span style="color:var(--text-primary)">' + esc(s) + '</span>'
+          + '</div>';
+      }
+    }
+  } catch (_) { /* defensive */ }
+  // 3. Always-available memory_search insert
+  html += '<div style="padding:6px 12px;font-size:10px;font-weight:600;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.04em;background:var(--bg-tertiary);margin-top:4px">Patterns</div>';
+  html += '<div onclick="sbInsertAtCursor(\\x27Use memory_search with query: \\\\\\\"YOUR_QUERY\\\\\\\".\\x27)" style="padding:7px 12px;cursor:pointer;font-size:12px;border-bottom:1px solid var(--border-light)" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27">'
+    + '<span style="font-family:ui-monospace,SFMono-Regular,Menlo,monospace;color:var(--text-primary)">memory_search</span>'
+    + '<span style="color:var(--text-muted);font-size:10px;margin-left:8px">Recall facts/transcripts at runtime</span>'
+    + '</div>';
+  html += '<div onclick="sbInsertAtCursor(\\x27Use memory_write to save this fact: \\\\\\\"FACT\\\\\\\" with reason \\\\\\\"WHY_IT_MATTERS\\\\\\\".\\x27)" style="padding:7px 12px;cursor:pointer;font-size:12px;border-bottom:1px solid var(--border-light)" onmouseover="this.style.background=\\x27var(--bg-tertiary)\\x27" onmouseout="this.style.background=\\x27transparent\\x27">'
+    + '<span style="font-family:ui-monospace,SFMono-Regular,Menlo,monospace;color:var(--text-primary)">memory_write</span>'
+    + '<span style="color:var(--text-muted);font-size:10px;margin-left:8px">Persist a new fact</span>'
+    + '</div>';
+  if (!html.trim()) html = '<div style="padding:18px;color:var(--text-muted);font-size:11px;text-align:center">No memory items surfaced.</div>';
+  listEl.innerHTML = html;
+}
+
 function sbInsertAtCursor(text) {
   var ed = document.getElementById('sb-editor');
   if (!ed) return;
@@ -28518,6 +28733,38 @@ function sbInsertAtCursor(text) {
   sbOnEdit();
 }
 
+// 1.18.132 — Test runner. Fires the open skill once via the same
+// /api/cron/run/:name path that the Skills detail Run-now button
+// uses (which itself works for unscheduled skills via cmdCronRun's
+// catalog fallback). Refuses to run if the file has unsaved changes
+// (otherwise the user would be testing the saved version, not what
+// they see in the editor — confusing).
+async function sbRunSkillTest() {
+  if (window._sbState.dirty) {
+    if (!confirm('Save current changes before testing? Unsaved edits won\\x27t be in the run.')) return;
+    await sbSaveCurrent();
+    if (window._sbState.dirty) return; // save failed
+  }
+  var name = window._sbState.skillName;
+  if (!name) return;
+  var btn = document.getElementById('sb-test-btn');
+  if (btn) { btn.disabled = true; btn.textContent = '⏳ Running…'; btn.style.color = 'var(--text-muted)'; btn.style.borderColor = 'var(--border)'; }
+  try {
+    var r = await window.apiFetch('/api/cron/run/' + encodeURIComponent(name), { method: 'POST' });
+    if (r.status === 409) {
+      toast('Already running. Wait for the in-flight run to finish.', 'warn');
+      return;
+    }
+    var d = await r.json();
+    if (!r.ok) { toast(d.error || 'Run failed', 'error'); return; }
+    toast('Started "' + name + '" — output streams to chat. Close the builder to see it.', 'success');
+  } catch (err) {
+    toast('Failed: ' + err, 'error');
+  } finally {
+    if (btn) { btn.disabled = false; btn.textContent = '▶ Test run'; btn.style.color = 'var(--green)'; btn.style.borderColor = 'var(--green)'; }
+  }
+}
+
 async function _openSkillModal(opts) {
   opts = opts || {};
   var existing = null;

package/dist/index.js

@@ -754,11 +754,14 @@ async function asyncMain() {
     // output to their Discord channel.
     const { AgentHeartbeatManager } = await import('./gateway/agent-heartbeat-manager.js');
     const agentHeartbeats = new AgentHeartbeatManager(gateway.getAgentManager(), gateway);
-    // Self-improve loop — closes the gap between "trigger written" and
-    // "fix applied." Every 10 min, scans self-improve/triggers/, classifies
-    // failures, auto-applies safe cron-config fixes, escalates risky ones.
-    const { SelfImproveLoop } = await import('./agent/self-improve-loop.js');
-    const selfImproveLoop = new SelfImproveLoop(dispatcher);
+    // Failure-fix consumer (1.18.134 — renamed from "self-improve-loop"
+    // to disambiguate from the Karpathy autoresearch SelfImproveLoop in
+    // src/agent/self-improve.ts). Every 10 min, scans
+    // self-improve/triggers/, classifies failures, auto-applies safe
+    // cron-config fixes, escalates risky ones. Different concern from
+    // the autoresearch hypothesize/evaluate loop.
+    const { FailureFixConsumer } = await import('./agent/failure-fix-consumer.js');
+    const failureFixConsumer = new FailureFixConsumer(dispatcher);
     // ── Build channel tasks ──────────────────────────────────────────
     const channelTasks = [];
     const activeChannels = [];
@@ -856,7 +859,38 @@ async function asyncMain() {
     heartbeat.start();
     cronScheduler.start();
     agentHeartbeats.start();
-    selfImproveLoop.start();
+    failureFixConsumer.start();
+    // 1.18.134 — nightly Karpathy-autoresearch self-improve trigger.
+    // The Karpathy SelfImproveLoop (src/agent/self-improve.ts) was
+    // previously only triggered by /self-improve run or CLI. With no
+    // automatic schedule the loop ran ~3 times in the prior 4 days and
+    // sat plateaued. This wires a daily 3am trigger so it iterates on
+    // its own — matching Karpathy's continuous-iteration model.
+    // SELF_IMPROVE_HOUR env var overrides (0–23, default 3).
+    const selfImproveHour = (() => {
+        const raw = parseInt(process.env.SELF_IMPROVE_HOUR ?? '3', 10);
+        if (Number.isFinite(raw) && raw >= 0 && raw <= 23)
+            return raw;
+        return 3;
+    })();
+    // node-cron is already a dependency (used by cron-scheduler). Schedule
+    // a single daily tick — the SelfImproveLoop's own time/iteration caps
+    // and plateau detection bound the work; we don't need finer granularity.
+    try {
+        const nodeCron = (await import('node-cron')).default;
+        nodeCron.schedule(`0 ${selfImproveHour} * * *`, () => {
+            logger.info({ hour: selfImproveHour }, 'Nightly self-improve trigger firing');
+            gateway.handleSelfImprove('run').then((summary) => {
+                logger.info({ summary }, 'Nightly self-improve trigger complete');
+            }).catch((err) => {
+                logger.error({ err }, 'Nightly self-improve trigger failed');
+            });
+        }, { timezone: process.env.TZ || 'America/Los_Angeles' });
+        logger.info({ hour: selfImproveHour }, `Self-improve nightly trigger scheduled (${selfImproveHour}:00 daily)`);
+    }
+    catch (err) {
+        logger.warn({ err }, 'Failed to schedule nightly self-improve trigger');
+    }
     // Background-task hygiene: any task left in 'running' is from a prior
     // process. Mark them aborted so the lifecycle is honest. (P6b will add
     // resumability; for now fail-fast is clearer than silently re-running.)
@@ -1086,7 +1120,7 @@ async function asyncMain() {
     heartbeat.stop();
     cronScheduler.stop();
     agentHeartbeats.stop();
-    selfImproveLoop.stop();
+    failureFixConsumer.stop();
     // ── Self-restart (enhanced with health check + rollback) ────────
     if (restartRequested) {
         // Clear our PID file BEFORE spawning the child, so ensureSingleton()

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "clementine-agent",
-  "version": "1.18.132",
+  "version": "1.18.134",
   "description": "Clementine — Personal AI Assistant (TypeScript)",
   "type": "module",
   "main": "dist/index.js",