sneakoscope 3.0.4 → 3.1.1

package/dist/core/zellij/zellij-slot-telemetry.js

@@ -1,6 +1,7 @@
 import path from 'node:path';
 import fsp from 'node:fs/promises';
-import { appendJsonlBounded, ensureDir, nowIso, readJson, readText } from '../fsx.js';
+import { appendJsonlBounded, ensureDir, nowIso, readJson, readText, writeTextAtomic } from '../fsx.js';
+import { withFileLock } from '../locks/file-lock.js';
 export const ZELLIJ_SLOT_TELEMETRY_EVENT_SCHEMA = 'sks.zellij-slot-telemetry-event.v1';
 export const ZELLIJ_SLOT_TELEMETRY_SNAPSHOT_SCHEMA = 'sks.zellij-slot-telemetry-snapshot.v1';
 const telemetrySnapshotCache = new Map();
@@ -71,7 +72,7 @@ export function mergeTelemetrySnapshots(base, overlay, opts = {}) {
     const slots = { ...(base.slots || {}) };
     for (const [key, row] of Object.entries(overlay.slots || {})) {
         const existing = slots[key];
-        slots[key] = !existing || telemetryTsMs(row.latest_ts) >= telemetryTsMs(existing.latest_ts) ? row : existing;
+        slots[key] = !existing ? row : newerSlotTelemetry(existing, row);
     }
     const baseTs = Date.parse(String(base.updated_at || '')) || 0;
     const overlayTs = Date.parse(String(overlay.updated_at || '')) || 0;
@@ -88,6 +89,21 @@ function telemetryTsMs(value) {
     const parsed = Date.parse(String(value || ''));
     return Number.isFinite(parsed) ? parsed : 0;
 }
+function newerSlotTelemetry(current, incoming) {
+    const currentTs = telemetryTsMs(current.latest_ts);
+    const incomingTs = telemetryTsMs(incoming.latest_ts);
+    const currentTerminal = isTelemetryTerminalStatus(current.status);
+    const incomingTerminal = isTelemetryTerminalStatus(incoming.status);
+    if (currentTerminal && !incomingTerminal)
+        return current;
+    if (!currentTerminal && incomingTerminal)
+        return incoming;
+    if (incomingTs > currentTs)
+        return incoming;
+    if (incomingTs < currentTs)
+        return current;
+    return incoming;
+}
 async function statTelemetryFile(file) {
     try {
         const st = await fsp.stat(file);
@@ -99,17 +115,17 @@ async function statTelemetryFile(file) {
 }
 export function applyTelemetryEventToSnapshot(snapshot, event) {
     const key = slotTelemetryKey(event.slot_id || event.worker_id, event.generation_index);
-    const slots = {
-        ...(snapshot.slots || {}),
-        [key]: mergeSlotTelemetry(snapshot.slots?.[key], event)
-    };
+    const previous = snapshot.slots?.[key];
+    const nextSlot = mergeSlotTelemetry(previous, event);
+    const slots = snapshot.slots || {};
+    slots[key] = nextSlot;
     return {
         schema: ZELLIJ_SLOT_TELEMETRY_SNAPSHOT_SCHEMA,
         mission_id: event.mission_id || snapshot.mission_id,
         updated_at: nowIso(),
         flush_count: snapshot.flush_count || 0,
         slots,
-        counts: countSlotTelemetry(slots)
+        counts: adjustSlotTelemetryCounts(snapshot.counts || countSlotTelemetry(snapshot.slots || {}), previous, nextSlot)
     };
 }
 export async function rebuildZellijSlotTelemetrySnapshot(root, missionId) {
@@ -176,11 +192,19 @@ function normalizeTelemetryEvent(event) {
     };
 }
 function mergeSlotTelemetry(previous, event) {
+    const previousTs = telemetryTsMs(previous?.latest_ts);
+    const eventTs = telemetryTsMs(event.ts);
+    const previousTerminal = isTelemetryTerminalStatus(previous?.status);
+    const eventTerminal = isTelemetryTerminalStatus(event.status);
+    const terminalRegression = Boolean(previous && previousTerminal && !eventTerminal);
+    const stale = Boolean(previous && (terminalRegression
+        || (!eventTerminal && eventTs < previousTs)
+        || (previousTerminal && eventTerminal && eventTs < previousTs)));
     return {
-        slot_id: event.slot_id,
-        generation_index: event.generation_index,
-        worker_id: event.worker_id,
-        status: event.status,
+        slot_id: stale ? previous.slot_id : event.slot_id,
+        generation_index: stale ? previous.generation_index : event.generation_index,
+        worker_id: stale ? previous.worker_id : event.worker_id,
+        status: stale ? previous.status : event.status,
         role: event.role || previous?.role || 'worker',
         backend: event.backend || previous?.backend || 'unknown',
         provider: event.provider || previous?.provider || 'unknown',
@@ -188,34 +212,55 @@ function mergeSlotTelemetry(previous, event) {
         worktree_id: event.worktree_id ?? previous?.worktree_id ?? null,
         worktree_path: event.worktree_path ?? previous?.worktree_path ?? null,
         task_title: event.task_title || previous?.task_title || 'waiting for task',
-        current_file: event.current_file ?? previous?.current_file ?? null,
-        latest_event_type: event.event_type,
-        latest_ts: event.ts,
-        progress: event.progress || previous?.progress || null,
+        current_file: stale ? previous.current_file ?? (terminalRegression ? null : event.current_file ?? null) : event.current_file ?? previous?.current_file ?? null,
+        latest_event_type: stale ? previous.latest_event_type : event.event_type,
+        latest_ts: stale ? previous.latest_ts : event.ts,
+        progress: stale ? previous.progress || (terminalRegression ? null : event.progress || null) : event.progress || previous?.progress || null,
         artifact_paths: unique([...(previous?.artifact_paths || []), ...(event.artifact_paths || [])]),
         blockers: unique([...(previous?.blockers || []), ...(event.blockers || [])]),
-        log_tail: event.log_tail || previous?.log_tail || ''
+        log_tail: stale ? previous.log_tail || (terminalRegression ? '' : event.log_tail || '') : event.log_tail || previous?.log_tail || ''
     };
 }
 function countSlotTelemetry(slots) {
     const counts = { queued: 0, running: 0, verifying: 0, completed: 0, failed: 0, headless: 0 };
     for (const row of Object.values(slots)) {
-        const status = normalizeStatus(row.status);
-        if (status === 'queued' || status === 'launching')
-            counts.queued += 1;
-        else if (status === 'running')
-            counts.running += 1;
-        else if (status === 'verifying')
-            counts.verifying += 1;
-        else if (status === 'completed' || status === 'drained')
-            counts.completed += 1;
-        else if (status === 'failed')
-            counts.failed += 1;
-        else if (status === 'headless')
-            counts.headless += 1;
+        const bucket = slotTelemetryCountBucket(row.status);
+        counts[bucket] += 1;
+    }
+    return counts;
+}
+function adjustSlotTelemetryCounts(current, previous, next) {
+    const counts = { ...current };
+    const nextBucket = slotTelemetryCountBucket(next.status);
+    if (previous) {
+        const previousBucket = slotTelemetryCountBucket(previous.status);
+        if (previousBucket !== nextBucket) {
+            counts[previousBucket] = Math.max(0, Number(counts[previousBucket] || 0) - 1);
+            counts[nextBucket] = Number(counts[nextBucket] || 0) + 1;
+        }
+        return counts;
     }
+    counts[nextBucket] = Number(counts[nextBucket] || 0) + 1;
     return counts;
 }
+function slotTelemetryCountBucket(status) {
+    const normalized = normalizeStatus(status);
+    if (normalized === 'queued' || normalized === 'launching')
+        return 'queued';
+    if (normalized === 'verifying')
+        return 'verifying';
+    if (normalized === 'completed' || normalized === 'drained')
+        return 'completed';
+    if (normalized === 'failed')
+        return 'failed';
+    if (normalized === 'headless')
+        return 'headless';
+    return 'running';
+}
+function isTelemetryTerminalStatus(status) {
+    const normalized = normalizeStatus(status);
+    return normalized === 'completed' || normalized === 'failed' || normalized === 'drained';
+}
 function normalizeStatus(value) {
     const text = String(value || '').toLowerCase();
     if (text === 'closed' || text === 'done' || text === 'passed')
@@ -258,20 +303,26 @@ function tail(value, max) {
 }
 async function writeTelemetrySnapshotFast(file, snapshot) {
     await ensureDir(path.dirname(file));
-    const flushCount = Number(telemetrySnapshotFlushCounts.get(file) || 0) + 1;
-    telemetrySnapshotFlushCounts.set(file, flushCount);
-    telemetrySnapshotLastFlushMs.set(file, Date.now());
-    // Merge with the on-disk snapshot before overwriting: multiple processes
-    // (orchestrator + worker children) flush this file concurrently and a plain
-    // overwrite would drop slots that only the other process has observed.
-    const disk = await readJson(file, null);
-    const merged = disk?.schema === ZELLIJ_SLOT_TELEMETRY_SNAPSHOT_SCHEMA ? mergeTelemetrySnapshots(disk, snapshot) : snapshot;
-    const next = { ...merged, flush_count: Math.max(flushCount, Number(merged.flush_count || 0)) };
-    telemetrySnapshotCache.set(file, next);
-    await fsp.writeFile(file, `${JSON.stringify(next)}\n`, 'utf8');
-    const stat = await statTelemetryFile(file);
-    if (stat)
-        telemetrySnapshotDiskStat.set(file, stat);
+    await withFileLock({
+        lockPath: `${file}.lock`,
+        timeoutMs: 30000,
+        staleMs: 2 * 60 * 1000
+    }, async () => {
+        const flushCount = Number(telemetrySnapshotFlushCounts.get(file) || 0) + 1;
+        telemetrySnapshotFlushCounts.set(file, flushCount);
+        telemetrySnapshotLastFlushMs.set(file, Date.now());
+        // Multiple processes (orchestrator + worker children) flush this file.
+        // Serialize the read/merge/write critical section so a slower writer
+        // cannot overwrite a newer slot observed by a different process.
+        const disk = await readJson(file, null);
+        const merged = disk?.schema === ZELLIJ_SLOT_TELEMETRY_SNAPSHOT_SCHEMA ? mergeTelemetrySnapshots(disk, snapshot) : snapshot;
+        const next = { ...merged, flush_count: Math.max(flushCount, Number(merged.flush_count || 0)) };
+        telemetrySnapshotCache.set(file, next);
+        await writeTextAtomic(file, `${JSON.stringify(next)}\n`);
+        const stat = await statTelemetryFile(file);
+        if (stat)
+            telemetrySnapshotDiskStat.set(file, stat);
+    });
 }
 function shouldFlushTelemetrySnapshot(file, event) {
     const next = (telemetrySnapshotWriteCounts.get(file) || 0) + 1;
@@ -289,7 +340,8 @@ function shouldFlushTelemetrySnapshot(file, event) {
         || event.event_type === 'worker_completed'
         || event.event_type === 'worker_failed'
         || event.status === 'completed'
-        || event.status === 'failed';
+        || event.status === 'failed'
+        || event.status === 'drained';
     const should = next === 1
         || important
         || now - last >= flushMs

package/dist/core/zellij/zellij-worker-pane-manager.js

@@ -14,12 +14,49 @@ export const ZELLIJ_PANE_CREATION_LOCK_METRICS_SCHEMA = 'sks.zellij-pane-creatio
 export function buildWorkerPaneName(slotId, generationIndex) {
     return `${slotId}/gen-${Math.max(1, Math.floor(Number(generationIndex) || 1))}`;
 }
-export function buildWorkerPaneTitle(slotId, generationIndex, context, serviceTier, backend, status, worktree) {
+export function buildWorkerPaneTitle(slotId, generationIndex, context, serviceTier, backend, status, worktree, taskTitle) {
+    // When a task label is available, the pane title is JUST the label + status:
+    // users identify panes by what they are doing, not by slot ids. Slot/gen,
+    // backend, provider, and worktree details stay inside the pane body and the
+    // SLOTS anchor rows. The verbose format remains the fallback for callers
+    // that have no task context.
+    const task = shortWorkerTaskLabel(taskTitle);
+    if (task)
+        return `${task} · ${workerBackendTag(backend, context?.provider)} · ${status || 'launching'}`;
     const base = buildWorkerPaneName(slotId, generationIndex);
     const normalized = normalizePaneProviderContext(context, serviceTier);
     const wt = worktree ? ` · WT:${worktree.id} · branch:${worktree.branch}` : '';
     return `${base}${wt} · ${backend || 'codex-sdk'} · ${providerPaneLabel(normalized)} · ${status || 'launching'}`;
 }
+// Local-LLM workers must be visually distinguishable from GPT/codex workers:
+// only trivial long-running edits belong on local models, while web research
+// and code review must run on GPT. The tag makes a misrouted task obvious.
+export function workerBackendTag(backend, provider) {
+    const text = `${String(backend || '')} ${String(provider || '')}`.toLowerCase();
+    if (/ollama|local/.test(text))
+        return 'LOCAL';
+    if (/fake|fixture/.test(text))
+        return 'fixture';
+    return 'GPT';
+}
+// Lead the pane title with a 1-3 word task label so users can tell WHAT each
+// slot is doing at a glance instead of an anonymous "slot-002/gen-1".
+export function shortWorkerTaskLabel(value, maxChars = 24) {
+    const text = String(value || '').replace(/\s+/g, ' ').trim();
+    if (!text)
+        return '';
+    const words = text.split(' ');
+    let label = '';
+    for (const word of words.slice(0, 3)) {
+        const next = label ? `${label} ${word}` : word;
+        if (next.length > maxChars)
+            break;
+        label = next;
+    }
+    if (!label)
+        label = text.slice(0, maxChars);
+    return label;
+}
 export function isRealZellijWorkerPaneIdSource(value) {
     return value === 'zellij_worker_new_pane_stdout' || value === 'zellij_worker_list_panes';
 }
@@ -162,7 +199,7 @@ export async function openWorkerPane(input) {
         });
         return record;
     }
-    const paneName = buildWorkerPaneTitle(input.slotId, input.generationIndex, providerContext, input.serviceTier, input.backend, input.statusLabel || 'running', input.worktree || null);
+    const paneName = buildWorkerPaneTitle(input.slotId, input.generationIndex, providerContext, input.serviceTier, input.backend, input.statusLabel || 'running', input.worktree || null, input.taskTitle || null);
     // CRITICAL: serialize anchor + worker pane creation per session. Workers
     // launch concurrently, and without this lock every worker raced past the
     // "does the SLOTS anchor exist yet?" check and created its OWN anchor with
@@ -695,7 +732,8 @@ async function withZellijPaneCreationLock(input, fn) {
     const current = new Promise((resolve) => {
         release = resolve;
     });
-    zellijPaneCreationLocks.set(key, previous.then(() => current, () => current));
+    const chained = previous.then(() => current, () => current);
+    zellijPaneCreationLocks.set(key, chained);
     await previous.catch(() => undefined);
     const acquiredAt = nowIso();
     const acquiredMs = Date.now();
@@ -741,7 +779,7 @@ async function withZellijPaneCreationLock(input, fn) {
             meta: { wait_ms: metrics.wait_ms, held_ms: metrics.held_ms }
         }).catch(() => undefined);
         release();
-        if (zellijPaneCreationLocks.get(key) === current)
+        if (zellijPaneCreationLocks.get(key) === chained)
             zellijPaneCreationLocks.delete(key);
     }
 }