claude-code-session-manager 0.75.1 → 0.75.3

package/src/main/scheduler.cjs

@@ -100,7 +100,7 @@ const queueOps = require('./queueOps.cjs');
 // home-dir layout.
 const { resolvePrdsDirs, resolveArchivedPrdsDirs, resolvePrdWriteDir, listEpicPrdDirs, listArchivedPrdDirs } = require('./lib/prdLocations.cjs');
 const { ensureEpic, appendPrdCreatedEvent, readActiveIndex } = require('./lib/epicMint.cjs');
-const { transitionJob } = require('./lib/scheduleJobTransitions.cjs');
+const { transitionJob, STATUS_HISTORY_CAP } = require('./lib/scheduleJobTransitions.cjs');
 const { buildContextDigest, composeExecutorPrompt } = require('./lib/epicContextDigest.cjs');
 const { JOB_STATUSES } = require('./lib/scheduleJobSchema.cjs');
 const { appendAuditEvent } = require('./lib/auditLog.cjs');
@@ -1424,7 +1424,7 @@ async function reconcile(state) {
         seen.add(job.slug);
         next.push({ ...job });
         console.warn(`[scheduler] reconcile: keeping ${job.status} job ${job.slug} — PRD source not visible in any candidate dir`);
-      } else if (job.status === 'completed' || job.status === 'failed') {
+      } else if (job.status === 'completed' || job.status === 'failed' || job.status === 'skipped') {
         terminalDroppedNeedingHistoryCheck.push(job);
       }
       continue;
@@ -2104,7 +2104,7 @@ async function clearPause(source) {
  * already-'completed' job, and that path is exactly what this guards.
  */
 function resetJobFields(job, errorMsg, opts = {}) {
-  if (job.status === 'completed' && opts.force !== true) return false;
+  if ((job.status === 'completed' || job.status === 'skipped') && opts.force !== true) return false;
   if (!transitionJob(job, 'pending', { reason: errorMsg ?? 'reset to pending', source: opts.source ?? 'resetJobFields' })) return false;
   job.runId = null;
   job.startedAt = null;
@@ -3486,13 +3486,18 @@ async function spawnJob(job, runId, runDir, defaultCwd) {
     // Stale queue entry: the PRD was archived (already shipped) or is gone
     // from disk entirely (deleted before dispatch). Both are retire-not-fail
     // results from prdArchivedSkipResult with exitCode 0 and no transcript —
-    // treat any truthy skip reason as a plain completion: no verify pass, no
-    // commit guard, no RCA feedback item.
+    // no verify pass, no commit guard, no RCA feedback item — but only
+    // 'prd-archived' actually means the work shipped (archivedTwinExists
+    // found the file under a prds-archived/ dir). 'prd-missing' means no
+    // executor ever ran at all, so it must NOT land on the same 'completed'
+    // status a genuine successful run produces — that conflation is what let
+    // 20 never-executed sigma PRDs read as shipped (2026-08-30 incident).
     if (res.skipped) {
+      const targetStatus = res.skipped === 'prd-missing' ? 'skipped' : 'completed';
       await mutate((s) => {
         const idx = s.jobs.findIndex((x) => x.slug === job.slug);
         if (idx >= 0) {
-          transitionJob(s.jobs[idx], 'completed', { reason: res.note ?? 'PRD archived or missing — treated as already-shipped', source: 'spawnJob:skip-archived' });
+          transitionJob(s.jobs[idx], targetStatus, { reason: res.note ?? 'PRD archived or missing — treated as already-shipped', source: 'spawnJob:skip-archived' });
           s.jobs[idx].finishedAt = new Date().toISOString();
           s.jobs[idx].exitCode = 0;
           s.jobs[idx].error = null;
@@ -4297,7 +4302,7 @@ async function pollLoop() {
 // callers/tests that only pass `jobs` are unaffected.
 function selectHistoryJobs(jobs, limit, historyEntries = []) {
   const cap = Math.max(1, Math.min(500, Number.isFinite(limit) ? Math.floor(limit) : 50));
-  const hot = (Array.isArray(jobs) ? jobs : []).filter((j) => j && (j.status === 'completed' || j.status === 'failed'));
+  const hot = (Array.isArray(jobs) ? jobs : []).filter((j) => j && (j.status === 'completed' || j.status === 'failed' || j.status === 'skipped'));
   const seen = new Set(hot.map((j) => `${j.slug}|${j.runId ?? ''}`));
   const archived = (Array.isArray(historyEntries) ? historyEntries : []).filter((j) => {
     if (!j) return false;
@@ -4743,6 +4748,43 @@ async function reverifyNeedsReview() {
   return { rescanned: candidates.length, healed, leftForReview };
 }
 
+/**
+ * schedule:clear-queue's core state mutation, split out so it's testable
+ * without ipcMain/electron: every victim (non-running job whose slug is in
+ * `victimSlugs`) must leave a durable history.jsonl record before it
+ * disappears from queue.json — a manual clear used to just filter the row
+ * out with no trace (a job could vanish with zero audit trail; see the
+ * 2026-08-30 sigma `788-pr-sweep-final-gate` incident). Non-terminal victims
+ * (pending/investigating/needs_review/quarantined) are force-transitioned to
+ * 'skipped' — never 'completed', since a manually-cleared job did not ship.
+ * Victims already terminal (completed/failed/skipped) keep their real
+ * status; they just get a statusHistory note + a history row so they don't
+ * silently drop before aging past HISTORY_RETENTION_MS.
+ *
+ * Mutates `state.jobs` in place (drops the victims, same contract as
+ * reconcile()) and returns the job snapshots to append to history.jsonl.
+ */
+function applyClearQueueVictims(state, victimSlugs, archiveNoteBySlug, archiveDir) {
+  const historyEntries = [];
+  for (const job of state.jobs) {
+    if (!victimSlugs.has(job.slug)) continue;
+    const reason = archiveNoteBySlug?.get(job.slug) ?? `PRD archived to ${archiveDir}`;
+    if (job.status === 'completed' || job.status === 'failed' || job.status === 'skipped') {
+      // Already terminal — keep the real status, just record the clear.
+      const history = Array.isArray(job.statusHistory) ? job.statusHistory : [];
+      history.push({ from: job.status, to: job.status, reason, source: 'clear-queue', at: new Date().toISOString() });
+      while (history.length > STATUS_HISTORY_CAP) history.shift();
+      job.statusHistory = history;
+    } else {
+      transitionJob(job, 'skipped', { reason, source: 'clear-queue' });
+    }
+    if (!job.finishedAt) job.finishedAt = new Date().toISOString();
+    historyEntries.push({ ...job });
+  }
+  state.jobs = state.jobs.filter((j) => !victimSlugs.has(j.slug));
+  return historyEntries;
+}
+
 function registerScheduleHandlers() {
   ensureDirs();
   supervisor.registerHandlers();
@@ -4905,10 +4947,6 @@ function registerScheduleHandlers() {
     return { ok: true, kind: 'info', message };
   });
 
-  // Archive every non-running PRD and drop its entry from queue.json.
-  // Running entries are kept (would orphan an in-flight job). PRD files are
-  // moved (not deleted) to prds-archived/<ISO>/ so the user can recover them.
-  // Path containment is enforced — only files inside PRDS_DIR are moved.
   ipcMain.handle('schedule:clear-queue', async () => {
     ensureDirs();
     const ts = new Date().toISOString().replace(/[:.]/g, '-');
@@ -4920,6 +4958,7 @@ function registerScheduleHandlers() {
     }
     await fsp.mkdir(archiveDir, { recursive: true });
     let archived = 0;
+    const archiveNoteBySlug = new Map();
     for (const job of victims) {
       const srcDir = await findPrdDir(job.slug) ?? prdDirForCwd(job.cwd);
       const src = path.resolve(path.join(srcDir, `${job.slug}.md`));
@@ -4928,17 +4967,27 @@ function registerScheduleHandlers() {
       try {
         await fsp.rename(src, dst);
         archived++;
+        archiveNoteBySlug.set(job.slug, `PRD file archived to ${archiveDir}`);
       } catch (e) {
         // ENOENT: the .md is already gone (reconcile would drop it on next
         // read anyway). Either way, fall through and remove from queue.
         if (e?.code !== 'ENOENT') {
           logs.writeLine({ level: 'warn', scope: 'scheduler', message: 'clear-queue: rename failed', meta: { slug: job.slug, error: e?.message } });
         }
+        archiveNoteBySlug.set(job.slug, `PRD file already gone (ENOENT) when archiving to ${archiveDir}`);
       }
     }
     await mutate(async (s) => {
       const victimSlugs = new Set(victims.map((j) => j.slug));
-      s.jobs = s.jobs.filter((j) => !victimSlugs.has(j.slug));
+      const historyEntries = applyClearQueueVictims(s, victimSlugs, archiveNoteBySlug, archiveDir);
+      // Append BEFORE this mutate's writeQueue persists s.jobs without the
+      // victims (mutate() only writes after this callback resolves) — same
+      // append-before-drop ordering queueHistory.cjs's reconcile() callers
+      // use, so a crash right after this handler returns can never leave a
+      // row gone from queue.json with nothing in history.jsonl yet.
+      if (historyEntries.length > 0) {
+        await queueHistory.appendHistory(historyEntries);
+      }
       await reconcile(s);
       return null;
     });
@@ -5353,6 +5402,20 @@ async function init() {
   }
 }
 
+// Measured against this repo's own live+archived PRD set (868 entries
+// across every local project — the admin route has no cwd filter by
+// default): no single field dominates a listPrdsInternal() entry: title is
+// the largest at ~30% of payload bytes, followed by slug/cwd/sourcePromptId/
+// epicId at 12-14% each — the 353KB the route was shipping came from
+// returning the full unbounded row count, not from one oversized field.
+// This trims the fields an agent doesn't need to answer "what PRDs are
+// there" (parallelGroup/estimateMinutes/sourcePromptId/epicId/
+// archivedStatus); `fields=full` restores them.
+function toCompactPrdEntry(entry) {
+  const { slug, title, cwd, mtimeMs, archived, status } = entry;
+  return { slug, title, cwd, mtimeMs, archived, status };
+}
+
 /**
  * listPrdsInternal() → every live + archived PRD across every project,
  * with each entry's real job status folded in (`status`: the live queue
@@ -5612,15 +5675,25 @@ const remote = {
 
   // Every live+archived PRD across every project (listPrdsInternal, shared
   // with the renderer's schedule:list-prds IPC handler), filtered by the
-  // admin route's cwd/epicId/status query params.
+  // admin route's cwd/epicId/status query params and paged by
+  // limit/offset. Sort order is listPrdsInternal's own slug.localeCompare
+  // (numeric) — stable across calls since it's a pure string sort, so
+  // paging by offset can't skip or duplicate an entry between requests as
+  // long as the underlying PRD set doesn't change mid-page.
   async listPrds(filter = {}) {
     const all = await listPrdsInternal();
-    return all.filter((entry) => {
+    const filtered = all.filter((entry) => {
       if (filter.cwd && entry.cwd !== filter.cwd) return false;
       if (filter.epicId && entry.epicId !== filter.epicId) return false;
       if (filter.status && entry.status !== filter.status) return false;
       return true;
     });
+    const total = filtered.length;
+    const limit = filter.limit ?? 100;
+    const offset = filter.offset ?? 0;
+    const page = filtered.slice(offset, offset + limit);
+    const prds = filter.fields === 'full' ? page : page.map(toCompactPrdEntry);
+    return { prds, total, limit, offset, hasMore: offset + page.length < total };
   },
 
   // Full body + parsed frontmatter for one PRD, live or archived. Mirrors
@@ -5759,7 +5832,7 @@ const remote = {
     const state = await readQueue();
     const job = state.jobs.find((j) => j.slug === slug);
     if (!job) return { ok: false, error: 'not found' };
-    if (job.status === 'completed' || job.status === 'failed' || job.status === 'needs_review') {
+    if (job.status === 'completed' || job.status === 'failed' || job.status === 'needs_review' || job.status === 'skipped') {
       return { ok: false, error: `job already terminal (status: "${job.status}") — nothing to cancel` };
     }
     const wasRunning = job.status === 'running';
@@ -5821,4 +5894,4 @@ function registerAdminRoutes(adminHttp, remoteObj = remote) {
   });
 }
 
-module.exports = { findOverrunningJobs, JOB_OVERRUN_FACTOR, JOB_OVERRUN_FLOOR_MS, registerScheduleHandlers, attachWindow, init, ROOT, PRDS_DIR, healRefusalReason, writeQueue, reconcile, reconcileSourcePromptId, allocateParallelGroup, selectHistoryJobs, parsePorcelain, FINISH_PROTOCOL, remote, pickNextBatch, pickForProject, reapDeadRunningJobs, pollRecoveryClearSource, memoryLimitedBatchSize, availableForJobs, reverifyNeedsReview, isRescanCandidate, isPromotableOriginal, selectAutoFixTargets, isEligibleForImmediateAutoFix, resolveRunId, isUnresolvableNeedsReview, isExhaustedAutoFix, isPlanUnqueued, fixSlugFor, healTargetForFix, buildInvestigationPrompt, isGitRepoSync, committedInWindow, computeCommittedDuringRun, classifySigtermWithCommit, isFixPlanSlug, isFixPlanBeyondDepthCap, MAX_INVESTIGATION_DEPTH, forceTickOutcome, applyPauseCleared, detectNetworkErrorInLog, detectRateLimitInLog, classifyFailureOutcome, commitGuardVerdict, TRANSIENT_RETRY_CAP, buildScheduleStatePayload, partitionBootOrphans, applyOrphanOutcome, BOOT_ORPHAN_KILL_GRACE_MS, registerAdminRoutes, notifyOriginatingTab, notifyNeedsReview, isNotifiableTerminalStatus, extractResultTextFromLog, candidatePrdsDirs, candidateArchivedPrdsDirs, resolveArchivedPrdStatus, prdDirForCwd, prdPathForJob, archivedPrdPathForJob, archivedTwinExists, findPrdDir, resolveVerifyPrdPath, resolveFixPlanPath, resolveNotifyPrd, runPrdMigration, consolidateAllFlatPrds, shouldSkipInvestigationForCleanRun, archiveCompletedPrd, retireCompletedSlugs, SCHEDULER_BOOTED_AT, SCHEDULER_CODE_SHA, resetJobFields, executeJob, prdArchivedSkipResult, spawnJob, listPrdsInternal, computeStallSummary, findStaleQuarantinedJobs, QUARANTINE_ESCALATE_MS };
+module.exports = { findOverrunningJobs, JOB_OVERRUN_FACTOR, JOB_OVERRUN_FLOOR_MS, registerScheduleHandlers, attachWindow, init, ROOT, PRDS_DIR, healRefusalReason, writeQueue, reconcile, reconcileSourcePromptId, allocateParallelGroup, selectHistoryJobs, parsePorcelain, FINISH_PROTOCOL, remote, pickNextBatch, pickForProject, reapDeadRunningJobs, pollRecoveryClearSource, memoryLimitedBatchSize, availableForJobs, reverifyNeedsReview, isRescanCandidate, isPromotableOriginal, selectAutoFixTargets, isEligibleForImmediateAutoFix, resolveRunId, isUnresolvableNeedsReview, isExhaustedAutoFix, isPlanUnqueued, fixSlugFor, healTargetForFix, buildInvestigationPrompt, isGitRepoSync, committedInWindow, computeCommittedDuringRun, classifySigtermWithCommit, isFixPlanSlug, isFixPlanBeyondDepthCap, MAX_INVESTIGATION_DEPTH, forceTickOutcome, applyPauseCleared, detectNetworkErrorInLog, detectRateLimitInLog, classifyFailureOutcome, commitGuardVerdict, TRANSIENT_RETRY_CAP, buildScheduleStatePayload, partitionBootOrphans, applyOrphanOutcome, BOOT_ORPHAN_KILL_GRACE_MS, registerAdminRoutes, notifyOriginatingTab, notifyNeedsReview, isNotifiableTerminalStatus, extractResultTextFromLog, candidatePrdsDirs, candidateArchivedPrdsDirs, resolveArchivedPrdStatus, prdDirForCwd, prdPathForJob, archivedPrdPathForJob, archivedTwinExists, findPrdDir, resolveVerifyPrdPath, resolveFixPlanPath, resolveNotifyPrd, runPrdMigration, consolidateAllFlatPrds, shouldSkipInvestigationForCleanRun, archiveCompletedPrd, retireCompletedSlugs, SCHEDULER_BOOTED_AT, SCHEDULER_CODE_SHA, resetJobFields, executeJob, prdArchivedSkipResult, spawnJob, listPrdsInternal, computeStallSummary, findStaleQuarantinedJobs, QUARANTINE_ESCALATE_MS, applyClearQueueVictims };

package/src/preload/api.d.ts

@@ -391,7 +391,7 @@ export interface ScheduleConfig {
  *  so this union is a manually-kept copy; drift from JOB_STATUSES (and from
  *  the renderer's own mirrors in StatusBadge.tsx/SchedulePanel.tsx) is
  *  caught by src/main/__tests__/scheduleJobStatusDrift.test.cjs. */
-export type ScheduleJobStatus = 'pending' | 'running' | 'investigating' | 'completed' | 'failed' | 'needs_review' | 'quarantined';
+export type ScheduleJobStatus = 'pending' | 'running' | 'investigating' | 'completed' | 'skipped' | 'failed' | 'needs_review' | 'quarantined';
 
 export interface ScheduleJobRuntime {
   pid: number;