karajan-code 2.19.2 → 2.19.4

package/README.md

@@ -345,6 +345,16 @@ Opt out: set `telemetry: false` in `~/.karajan/kj.config.yml`
 
 ## Recent releases
 
+> **v2.19.4 released** — Patch. **`kj resume` now continues from where it stopped** + **autoInit no longer commits zombies on the user's main**. Two bugs closed in one release.
+>
+> KJC-BUG-0058 (PR #798, reported by [@aitormf](https://github.com/aitormf)): a session that stopped during Sonar would re-run the full pre-loop on `kj resume <id>` — HU-reviewer, intent, discover, triage, domainCurator, **researcher, architect, planner** all from scratch — doubling token cost and breaking the value-prop of the command. Root cause: `resumeFlow` (flow-runner.js:280) called `runFlow` without rehydrating stage state, and the session never persisted stage outputs in the first place. **Fix**: two new mutators in `src/session/mutators.js` (`setStageResult` mirrors into `stage_results[name]` + `stages_completed[]`; `setStageBundle` adds `stage_bundles[name]` for cross-stage context like `researchContext`, `architectContext`, `plannedTask`). Two closures inside `runPreLoopStages` (`persistStage` + `resumeSkip`) wrap every cacheable stage. `init-context.js` rehydrates `ctx.stageResults` from the loaded session before invoking the pre-loop. Triage is *not* skipped on resume — it produces `roleOverrides` the Brain decisor depends on and is cheap to re-run.
+>
+> KJC-BUG-0060 (PR #797, reported by mjfosela during the v2.19.3 release): `git checkout main` reported `[ahead 27]` of origin/main. Every commit was titled `initial commit`, authored by `manufosela@gmail.com`, and had the **exact same tree as its parent** — completely empty. The reflog held **2 495 such SHAs** accumulated since April 2026. None ever reached origin/main (gh push / CI would have rejected them) so runtime impact was zero, but the local history was noisy and on every release it looked like a sync loss. Root cause: `src/orchestrator/config-init.js::autoInit()` guarded with `!(await exists(projectDir/.git))`, which fails two ways: (a) dogfooding kj on karajan-code itself from a subdir → `exists()` returns false → `git init` re-initializes the *parent's* `.git/` (idempotent, harmless) → `git commit --allow-empty` then resolves upward and lands an empty commit on the parent's main; (b) transient FS hicks (EACCES/ENOENT) flip `exists()` to a false negative. **Fix**: switch the static FS probe for `git rev-parse --is-inside-work-tree`, which performs the same upward search git would use for the commit itself — guard cannot disagree with the operation it guards. And drop the `git commit --allow-empty -m "initial commit"` step entirely: no downstream stage needs a root commit; the 2 495 zombies never broke anything, the seed was decorative and turned out to be the actual user-visible symptom.
+>
+> **If your `kj resume` re-runs researcher/architect, or your `git status` shows mysterious `[ahead N]` after dogfooding kj on a kj-linked source tree, upgrade to v2.19.4.**
+>
+> **v2.19.3 released** — Patch. **HU Board now reads + writes plans from the canonical home dir** (KJC-BUG-0059). Reported by [@aitormf](https://github.com/aitormf): the top card showed `Directorio del proyecto — no detectado` even when the run had a valid `projectDir` and the coder was reading files from it. Root cause: five board call sites still hard-coded `~/.kj/plans/` as their plans root — leftover from the v2.19.0 home consolidation, which fixed `sync.js` but missed the rest. After the auto-migrator ran (or the user created new plans post-v2.19.0), plans landed under `~/.karajan/plans/<slug>/`; the board kept looking under `~/.kj/plans/<slug>/` and silently found nothing — so `GET /api/projects/:id/preflight` could not extract `projectDir` (the literal Aitor saw), `GET /api/projects/:id/plans-outcome` returned `plans: []` for every project, `DELETE /api/projects/:id` swept the wrong path leaving residue on disk, `DELETE /api/plans/:planId` failed silently, `plan-mutations.plansRoot` wrote new per-HU run logs to the legacy root splitting state across both, and `cleanup-zombies` never GC'd zombies under `~/.karajan/plans/`. PR #795 ships three new exports in `packages/hu-board/src/db.js`: `getHuBoardPlansDir()` (canonical, or `KJ_PLANS_DIR` override), `getHuBoardLegacyPlansDir()` (legacy, null when override set), `getHuBoardPlansDirs()` ordered `[canonical, legacy?]` for read callers. Single-write callers (`plan-mutations`) use the canonical root; read / delete / GC iterate both so users mid-migration with plans still under `~/.kj/` don't regress. 29 hu-board test files / 349 tests still green. **If your board shows "Directorio del proyecto — no detectado" on a project that clearly works in CLI, upgrade to v2.19.3.**
+>
 > **v2.19.2 released** — Patch. **SonarQube 401 now triggers automatic token re-bootstrap** instead of failing the run (KJC-BUG-0057). Until v2.19.1, when the Sonar token was missing / stale / revoked / pointed at a recreated Sonar instance, `kj run` / `kj audit` threw `SonarQube authentication failed (HTTP 401)` with the hint "Regenerate with `kj init`" — putting the user in the loop for plumbing Karajan can do itself. PR #793 wires `src/sonar/api.js::sonarFetchOnce` to invoke a new `src/sonar/token-recovery.js::recoverSonarToken()` on the first 401, which reuses `bootstrapSonarToken()` (already shipped in v2.10.2) — probes admin/admin, rotates the default password if still in place, revokes the existing `karajan-cli` token, generates a fresh GLOBAL_ANALYSIS_TOKEN, mutates `config.sonarqube.token` in place AND mirrors it to `~/.karajan/sonar-credentials.json` so future processes pick it up via the normal resolver chain instead of triggering recovery again. The original request retries once with the new token; the user never sees the 401 when recovery succeeds. Per-process latch ensures N endpoints 401-ing trigger ONE bootstrap, not N. Programmatic, **zero LLM involvement** — exactly the kind of plumbing Karajan should never delegate to the agent. Reported by [@aitormf](https://github.com/aitormf).
 >
 > **v2.19.1 released** — Patch. **APPLICATION BLOCKER fix** for the HU Board: every fresh `npm install -g karajan-code` since the HU Board feature shipped was producing tarballs without the `packages/hu-board/` directory (the `files` array in `package.json` did not include it), and even when the user copied that directory manually, the board crashed at startup with `Cannot find package 'helmet' imported from .../packages/hu-board/src/server.js` because its five dependencies (`helmet`, `chokidar`, `better-sqlite3`, `express`, `express-rate-limit`) were declared in `packages/hu-board/package.json` but missing from the root `dependencies`. PR #791 adds `packages/hu-board/{src,public,package.json}` to `files`, adds the five deps at root at the exact same versions so `npm dedupe` collapses to one copy resolvable by upward traversal, and regenerates `package-lock.json`. Verified end-to-end: `npm pack --dry-run` now ships 12 board files; `node packages/hu-board/src/server.js` boots cleanly. Also internal: 38 direct `os.homedir()` callers routed through the unified resolver (PR #790, `KARAJAN_HOME` now redirects every component) and 5 inline constructions of `~/.karajan/hu-board-runs/` unified under one helper (PR #789). Reported by [@aitormf](https://github.com/aitormf). **If you tried v2.19.0 and `kj board start` failed, upgrade to v2.19.1.**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "karajan-code",
-  "version": "2.19.2",
+  "version": "2.19.4",
   "description": "Local multi-agent coding orchestrator with TDD, SonarQube, and code review pipeline",
   "type": "module",
   "license": "AGPL-3.0",

package/packages/hu-board/src/cleanup-zombies.js

@@ -47,8 +47,12 @@ function isEphemeralId(id) {
   return EPHEMERAL_PREFIXES.some((p) => id.startsWith(p));
 }
 
-function plansRoot() {
-  return process.env.KJ_PLANS_DIR || path.join(homedir(), '.kj', 'plans');
+// KJC-BUG-0059 (v2.19.3): cleanup walks all known plan roots,
+// canonical first then legacy. Pre-fix hard-coded `~/.kj/plans/`
+// missed every plan post-v2.19.0 → zombies never garbage-collected.
+import { getHuBoardPlansDirs } from './db.js';
+function plansRoots() {
+  return getHuBoardPlansDirs();
 }
 
 function olderThanDays(isoStr, days) {
@@ -90,7 +94,7 @@ export function cleanupZombies(opts = {}) {
     const fsPaths = [
       ...storyIds.map((sid) => path.join(getKjHome(), 'hu-stories', sid)),
       ...sessionIds.map((sid) => path.join(getKjHome(), 'sessions', sid)),
-      path.join(plansRoot(), proj.id),
+      ...plansRoots().map((root) => path.join(root, proj.id)),
     ];
     deleteProject(proj.id);
     addTombstone('project', proj.id, { source: 'auto-cleanup', fsPaths });

package/packages/hu-board/src/db.js

@@ -56,6 +56,51 @@ export function getHuBoardRunsDir() {
   return join(getKjHome(), 'hu-board-runs');
 }
 
+/**
+ * Canonical path to the plans directory the board scans
+ * (`<karajan-home>/plans/`). Respects `KJ_PLANS_DIR` as an absolute
+ * override, mirroring sync.js. KJC-BUG-0059 (v2.19.3): preflight
+ * endpoint used a hard-coded `~/.kj/plans/` legacy fallback that
+ * survived the v2.19.0 consolidation, causing the board to report
+ * "directorio del proyecto: no detectado" even when the plan on disk
+ * had a valid `projectDir`.
+ *
+ * @returns {string}
+ */
+export function getHuBoardPlansDir() {
+  return process.env.KJ_PLANS_DIR || join(getKjHome(), 'plans');
+}
+
+/**
+ * Legacy `~/.kj/plans/` directory — only returned when KJ_PLANS_DIR
+ * is unset (it would be the same path anyway when overridden).
+ * Callers iterate canonical + legacy so users who have not yet run
+ * the v2.19.0 auto-migrator still see their plans. KJC-BUG-0059.
+ *
+ * @returns {string|null}
+ */
+export function getHuBoardLegacyPlansDir() {
+  if (process.env.KJ_PLANS_DIR) return null;
+  return join(process.env.HOME || '/root', '.kj', 'plans');
+}
+
+/**
+ * Ordered list of every plans root the board should scan: canonical
+ * first (`~/.karajan/plans/`), legacy second when applicable
+ * (`~/.kj/plans/`). Honours `KJ_PLANS_DIR` as a single override,
+ * matching `sync.js`'s fullScan / startWatcher precedence so the
+ * preflight, plans-outcome, delete-project and delete-plan endpoints
+ * see the same set of files the watcher sees. KJC-BUG-0059.
+ *
+ * @returns {string[]}
+ */
+export function getHuBoardPlansDirs() {
+  const dirs = [getHuBoardPlansDir()];
+  const legacy = getHuBoardLegacyPlansDir();
+  if (legacy) dirs.push(legacy);
+  return dirs;
+}
+
 // Bump when the schema gains a column / index / table that older
 // Karajan versions cannot understand. A DB written by a newer Karajan
 // refuses to open here so it does not get silently downgraded.

package/packages/hu-board/src/plan-mutations.js

@@ -40,10 +40,13 @@ const KJ_CLI = join(REPO_ROOT, 'src', 'cli.js');
  * `KJ_PLANS_DIR` (tests) then `KJ_HOME/plans` (power users) then
  * `~/.kj/plans/` (default) — same precedence as `sync.js::fullScan`.
  */
+// KJC-BUG-0059: write operations go to the canonical plans dir.
+// Read operations should iterate getHuBoardPlansDirs() instead, to
+// pick up plans the user has under the legacy ~/.kj/plans/ tree
+// while the auto-migrator has not yet fired.
+import { getHuBoardPlansDir } from './db.js';
 function plansRoot() {
-  if (process.env.KJ_PLANS_DIR) return process.env.KJ_PLANS_DIR;
-  if (process.env.KJ_HOME) return join(process.env.KJ_HOME, 'plans');
-  return join(homedir(), '.kj', 'plans');
+  return getHuBoardPlansDir();
 }
 
 /**

package/packages/hu-board/src/preflight.js

@@ -30,6 +30,7 @@
 import { execSync } from 'node:child_process';
 import { existsSync, readFileSync, readdirSync, statSync } from 'node:fs';
 import { join } from 'node:path';
+import { getHuBoardPlansDirs } from './db.js';
 
 function runQuiet(cmd, opts = {}) {
   try {
@@ -221,13 +222,13 @@ function checkConfigYml() {
 }
 
 function checkPlans(projectId) {
-  // Plans live at ~/.kj/plans/<projectId>/*.json (or
-  // $KJ_HOME/plans/<projectId>/*.json under VITEST or custom config).
-  // Mirror plan-mutations.js::plansRoot() so the path matches reality.
-  const dir = process.env.KJ_HOME
-    ? join(process.env.KJ_HOME, 'plans', projectId)
-    : join(process.env.HOME || '', '.kj', 'plans', projectId);
-  if (!existsSync(dir)) {
+  // KJC-BUG-0059 (v2.19.3): scan canonical + legacy roots, same as
+  // the preflight route. Pre-fix this hard-coded `~/.kj/plans/`
+  // missed every plan post-v2.19.0 → preflight reported "sin plans
+  // aún" even when the user had a valid plan on disk.
+  const dirs = getHuBoardPlansDirs().map((root) => join(root, projectId));
+  const dir = dirs.find((d) => existsSync(d));
+  if (!dir) {
     return {
       id: 'plans', label: 'Plans en disco', status: 'info',
       detail: 'sin plans aún',

package/packages/hu-board/src/routes/api.js

@@ -15,6 +15,9 @@ import {
   deleteSession,
   getKjHome,
   getHuBoardRunsDir,
+  getHuBoardPlansDir,
+  getHuBoardLegacyPlansDir,
+  getHuBoardPlansDirs,
   getStoryRow,
   listPlanIdsForProject,
   updateStoryStatus,
@@ -192,22 +195,23 @@ router.get('/projects/:id/preflight', async (req, res) => {
   try {
     const projectId = req.params.id;
     let projectDir = null;
+    // KJC-BUG-0059 (v2.19.3): scan plans in BOTH the canonical
+    // ~/.karajan/plans/ AND the legacy ~/.kj/plans/ — the legacy
+    // fallback covers users who started the board before running any
+    // `kj` command on a fresh install (so the auto-migrator from
+    // v2.19.0 has not fired yet). Pre-fix, this endpoint hard-coded
+    // ~/.kj/plans/ as the default, missing every plan post-2.19.0
+    // → board card showed "directorio del proyecto: no detectado"
+    // even though the plan on disk had a valid projectDir.
     try {
-      // Plans live at ~/.kj/plans/<slug>/ when KJ_HOME is unset, or
-      // $KJ_HOME/plans/<slug>/ when set. db.js::getKjHome() returns
-      // the .karajan root, NOT the .kj root — they are sister dirs.
-      // Mirror plan-mutations.js::plansRoot() exactly.
-      const plansBase = process.env.KJ_HOME
-        ? path.join(process.env.KJ_HOME, 'plans')
-        : path.join(process.env.HOME || '', '.kj', 'plans');
-      const plansDir = path.join(plansBase, projectId);
-      if (fs.existsSync(plansDir)) {
-        const files = fs.readdirSync(plansDir).filter((f) => f.endsWith('.json'));
-        for (const f of files) {
+      outer: for (const root of getHuBoardPlansDirs()) {
+        const plansDir = path.join(root, projectId);
+        if (!fs.existsSync(plansDir)) continue;
+        for (const f of fs.readdirSync(plansDir).filter((x) => x.endsWith('.json'))) {
           try {
             const plan = JSON.parse(fs.readFileSync(path.join(plansDir, f), 'utf8'));
-            if (plan?.projectDir) { projectDir = plan.projectDir; break; }
-          } catch { /* try next */ }
+            if (plan?.projectDir) { projectDir = plan.projectDir; break outer; }
+          } catch { /* try next file */ }
         }
       }
     } catch { /* projectDir stays null → preflight will report it */ }
@@ -229,17 +233,17 @@ router.get('/projects/:id/preflight', async (req, res) => {
 router.get('/projects/:id/plans-outcome', (req, res) => {
   try {
     const projectId = req.params.id;
-    const plansBase = process.env.KJ_HOME
-      ? path.join(process.env.KJ_HOME, 'plans')
-      : path.join(process.env.HOME || '', '.kj', 'plans');
-    const plansDir = path.join(plansBase, projectId);
-    if (!fs.existsSync(plansDir)) return res.json({ projectId, plans: [] });
+    // KJC-BUG-0059 (v2.19.3): scan canonical + legacy roots so users
+    // who have not yet run the auto-migrator still see their plans.
     const out = [];
-    for (const f of fs.readdirSync(plansDir).filter(n => n.endsWith('.json'))) {
-      try {
-        const plan = JSON.parse(fs.readFileSync(path.join(plansDir, f), 'utf8'));
-        if (plan?.version !== 2) continue;
-        out.push({
+    for (const root of getHuBoardPlansDirs()) {
+      const plansDir = path.join(root, projectId);
+      if (!fs.existsSync(plansDir)) continue;
+      for (const f of fs.readdirSync(plansDir).filter(n => n.endsWith('.json'))) {
+        try {
+          const plan = JSON.parse(fs.readFileSync(path.join(plansDir, f), 'utf8'));
+          if (plan?.version !== 2) continue;
+          out.push({
           planId: plan.planId,
           name: plan.name || (plan.task ? plan.task.slice(0, 80) : plan.planId),
           status: plan.status || 'draft',
@@ -248,6 +252,7 @@ router.get('/projects/:id/plans-outcome', (req, res) => {
           huCount: Array.isArray(plan.hus) ? plan.hus.length : 0,
         });
       } catch { /* skip unreadable plan */ }
+      }
     }
     res.json({ projectId, plans: out });
   } catch (err) {
@@ -356,12 +361,13 @@ router.delete('/projects/:id', (req, res) => {
     // ~/.kj/plans relative to KJ_HOME's parent, which broke when the
     // two homes were configured separately and left orphan plan dirs
     // on disk after a 🗑️ delete.
-    const plansRoot = process.env.KJ_PLANS_DIR
-      || path.join(os.homedir(), '.kj', 'plans');
+    // KJC-BUG-0059 (v2.19.3): when removing a deleted project's
+    // residual plan dirs, sweep BOTH the canonical and legacy roots
+    // so the cleanup is consistent for users mid-migration.
     const fsPaths = [
       ...storyIds.map((sid) => path.join(huStoriesDir(), sid)),
       ...sessionIds.map((sid) => path.join(getKjHome(), 'sessions', sid)),
-      path.join(plansRoot, id),
+      ...getHuBoardPlansDirs().map((root) => path.join(root, id)),
     ];
 
     const existed = deleteProject(id);
@@ -464,13 +470,18 @@ router.delete('/sessions/:id', (req, res) => {
 router.delete('/plans/:planId', (req, res) => {
   try {
     const planId = req.params.planId;
-    // Locate the plan file by scanning the plans dir for filename match.
-    const plansRoot = process.env.KJ_PLANS_DIR || path.join(getKjHome(), '..', '.kj', 'plans');
+    // Locate the plan file by scanning both canonical and legacy plans
+    // roots — KJC-BUG-0059 (v2.19.3). Pre-fix this hard-coded
+    // `getKjHome()/../.kj/plans` which silently misses plans saved by
+    // post-2.19.0 callers under `~/.karajan/plans/`.
     let foundPath = null;
     try {
-      for (const projSlug of fs.readdirSync(plansRoot)) {
-        const candidate = path.join(plansRoot, projSlug, `plan-${planId}.json`);
-        if (fs.existsSync(candidate)) { foundPath = candidate; break; }
+      outer: for (const plansRoot of getHuBoardPlansDirs()) {
+        if (!fs.existsSync(plansRoot)) continue;
+        for (const projSlug of fs.readdirSync(plansRoot)) {
+          const candidate = path.join(plansRoot, projSlug, `plan-${planId}.json`);
+          if (fs.existsSync(candidate)) { foundPath = candidate; break outer; }
+        }
       }
     } catch { /* plansRoot missing */ }
     addTombstone('plan', planId, { source: 'api', fsPaths: foundPath ? [foundPath] : [] });

package/src/orchestrator/config-init.js

@@ -26,17 +26,27 @@ import { exists, ensureDir } from "../utils/fs.js";
  * Called by the orchestrator before the pipeline starts.
  */
 export async function autoInit(projectDir, logger) {
-  // Ensure git repo exists — without git, diff/reviewer/commit won't work
-  const gitDir = path.join(projectDir, ".git");
-  if (!(await exists(gitDir))) {
-    // Audit follow-up: was execSync with constant strings. Inputs are
-    // hardcoded, so injection isn't possible, but the project standard
-    // since #555 is execFileSync everywhere. Migrated for consistency.
-    const { execFileSync } = await import("node:child_process");
+  // Ensure git repo exists — without git, diff/reviewer/commit won't work.
+  //
+  // KJC-BUG-0060: the previous guard `!exists(projectDir/.git)` was insufficient
+  // when projectDir is a subdirectory inside an existing repo (common in dogfooding
+  // scenarios where kj-linked runs from the karajan-code source tree). In that case
+  // `git commit --allow-empty` would resolve upward to the parent repo and land an
+  // empty "initial commit" on the user's main branch. Use `git rev-parse
+  // --is-inside-work-tree` instead — it follows the same upward traversal that git
+  // does naturally, so a subdir inside a repo correctly reports `true` and we don't
+  // touch anything. We also drop the `git commit --allow-empty -m "initial commit"`
+  // step: a root commit is not required for any downstream stage (diff, review,
+  // coder, sonar) and the empty-commit zombie history was the user-visible symptom.
+  const { execFileSync } = await import("node:child_process");
+  try {
+    execFileSync("git", ["rev-parse", "--is-inside-work-tree"], { cwd: projectDir, stdio: "pipe" });
+    // Already inside a git work tree (own .git or parent's). Nothing to do.
+  } catch {
+    // Not inside any repo — create a fresh one. No empty seed commit.
     try {
       execFileSync("git", ["init"], { cwd: projectDir, stdio: "pipe" });
-      execFileSync("git", ["commit", "--allow-empty", "-m", "initial commit"], { cwd: projectDir, stdio: "pipe" });
-      logger.info("Initialized git repository with empty initial commit");
+      logger.info("Initialized git repository (no seed commit)");
     } catch (err) {
       logger.warn(`Failed to init git repo: ${err.message}`);
     }

package/src/orchestrator/drivers/init-context.js

@@ -172,7 +172,13 @@ export async function initFlowContext({ task, config, logger, emitter, askQuesti
     })
   );
 
-  ctx.stageResults = {};
+  // KJC-BUG-0058: on `kj resume`, rehydrate stage results from the loaded
+  // session so completed pre-loop stages (researcher, architect, planner…)
+  // are skipped instead of re-executed. `setStageResult` mirrors writes
+  // into both `stage_results` (full payload) and `stages_completed` (flat
+  // array) — copying `stage_results` here is sufficient because
+  // `runPreLoopStages` only checks membership in `stageResults`.
+  ctx.stageResults = { ...(ctx.session?.stage_results || {}) };
   ctx.sonarState = { issuesInitial: null, issuesFinal: null };
 
   const preLoopResult = await runPreLoopStages({ config, logger, emitter, eventBase: ctx.eventBase, session: ctx.session, flags, pipelineFlags: ctx.pipelineFlags, coderRole: ctx.coderRole, trackBudget: ctx.trackBudget, task, askQuestion, pgTaskId, pgProject, stageResults: ctx.stageResults, brainCtx: ctx.brainCtx });

package/src/orchestrator/drivers/pre-loop.js

@@ -34,6 +34,7 @@ import { refineSkillsSemantically, resolveSkillsMode } from "../../skills/semant
 import { saveSession } from "../../session/store.js";
 import {
   setPreflight, setAutoInstalledSkills, setSkillsRecommended,
+  setStageResult, setStageBundle,
 } from "../../session/mutators.js";
 import {
   runResearcherStage, runArchitectStage, runPlannerStage,
@@ -60,18 +61,42 @@ import { tryCiComment } from "../ci-integration.js";
 import { getIntegration } from "../integrations.js";
 
 export async function runPreLoopStages({ config, logger, emitter, eventBase, session, flags, pipelineFlags, coderRole, trackBudget, task, askQuestion, pgTaskId, pgProject, stageResults, brainCtx }) {
+  // KJC-BUG-0058: on `kj resume` `stageResults` is rehydrated from
+  // `session.stage_results`. Persist after every successful pre-loop stage
+  // and skip the work when the slot is already populated. Cost: one
+  // `await saveSession` per stage; equivalent to the writes the post-loop
+  // path already makes for security / acceptance results.
+  const persistStage = async (name, result) => {
+    if (!result) return;
+    stageResults[name] = result;
+    setStageResult(session, name, result);
+    try { await saveSession(session); } catch (err) {
+      logger.warn(`Could not persist '${name}' stage result: ${err.message}`);
+    }
+  };
+  const resumeSkip = (name) => {
+    if (!stageResults[name]) return false;
+    logger.info(`[resume] skipping pre-loop stage '${name}' — completed in previous session`);
+    emitProgress(emitter, makeEvent("stage:skipped", { ...eventBase, stage: name }, {
+      status: "ok",
+      message: `Stage '${name}' skipped — already completed on previous run`,
+      detail: { cached: true }
+    }));
+    return true;
+  };
+
   // --- HU Reviewer (first stage, before everything else, opt-in) ---
   const huFile = flags.huFile || null;
   if (flags.enableHuReviewer !== undefined) pipelineFlags.huReviewerEnabled = Boolean(flags.enableHuReviewer);
-  if (pipelineFlags.huReviewerEnabled && huFile) {
+  if (pipelineFlags.huReviewerEnabled && huFile && !resumeSkip("huReviewer")) {
     const huResult = await runHuReviewerStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget, huFile, askQuestion, pgStories: null });
-    stageResults.huReviewer = huResult.stageResult;
+    await persistStage("huReviewer", huResult.stageResult);
   }
 
   // --- Intent classifier (deterministic pre-triage, opt-in) ---
-  if (config.guards?.intent?.enabled) {
+  if (config.guards?.intent?.enabled && !resumeSkip("intent")) {
     const intentResult = classifyIntent(task, config);
-    stageResults.intent = intentResult;
+    await persistStage("intent", intentResult);
     if (intentResult.classified) {
       emitProgress(emitter, makeEvent("intent:classified", { ...eventBase, stage: "intent" }, {
         message: `Intent classified: ${intentResult.taskType} (${intentResult.level}) — ${intentResult.message}`,
@@ -82,21 +107,21 @@ export async function runPreLoopStages({ config, logger, emitter, eventBase, ses
 
   // --- Discover (pre-triage, opt-in) ---
   if (flags.enableDiscover !== undefined) pipelineFlags.discoverEnabled = Boolean(flags.enableDiscover);
-  if (pipelineFlags.discoverEnabled) {
+  if (pipelineFlags.discoverEnabled && !resumeSkip("discover")) {
     const discoverResult = await runDiscoverStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
-    stageResults.discover = discoverResult.stageResult;
+    await persistStage("discover", discoverResult.stageResult);
   }
 
   // --- Triage (always on) — routed through StageRegistry (TSK-0336) ---
-  // canRun here is `pipelineFlags.triageEnabled !== false`; triageEnabled is
-  // never set to false in production, so behavior matches the previous
-  // unconditional call. If a caller ever flips it off, runStage returns null
-  // and we skip apply+persist with reasonable defaults (no overrides, no
-  // stageResult).
+  // KJC-BUG-0058: NOT skipped on resume. Triage is cheap and emits
+  // `roleOverrides` that downstream stages depend on (Brain decisor + the
+  // pipelineFlags). Re-running it on resume is the safe path; the heavy
+  // stages it gates (researcher, architect, planner) ARE skipped if
+  // already complete.
   const triageCtx = { config, logger, emitter, eventBase, session, coderRole, trackBudget, pipelineFlags };
   const triageResult = await runStage(stageRegistry.get("triage"), triageCtx) ?? { roleOverrides: null, stageResult: null };
   applyTriageOverrides(pipelineFlags, triageResult.roleOverrides);
-  stageResults.triage = triageResult.stageResult;
+  await persistStage("triage", triageResult.stageResult);
 
   // --- Brain decisor (opt-in, intent-driven routing) ---
   //
@@ -188,13 +213,13 @@ export async function runPreLoopStages({ config, logger, emitter, eventBase, ses
 
   // --- Domain Curator (after triage + skill auto-install, before planning phases) ---
   const domainHints = triageResult.stageResult?.domainHints || [];
-  if (domainHints.length > 0 || config.projectDir) {
+  if ((domainHints.length > 0 || config.projectDir) && !resumeSkip("domainCurator")) {
     try {
       const { domainContext, stageResult: dcStageResult } = await runDomainCuratorStage({
         config, logger, emitter, eventBase, session, trackBudget,
         domainHints, askQuestion
       });
-      stageResults.domainCurator = dcStageResult;
+      await persistStage("domainCurator", dcStageResult);
       if (domainContext) {
         config = { ...config, domainContext };
       }
@@ -213,7 +238,7 @@ export async function runPreLoopStages({ config, logger, emitter, eventBase, ses
       pgStories = getIntegration("tracker")?.buildHuStoriesFromCard?.(session.pg_card) ?? null;
     }
     const huResult = await runHuReviewerStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget, huFile: null, askQuestion, pgStories });
-    stageResults.huReviewer = huResult.stageResult;
+    await persistStage("huReviewer", huResult.stageResult);
   }
 
   // --- Auto-simplify pipeline for simple tasks (before explicit flag overrides) ---
@@ -291,7 +316,7 @@ export async function runPreLoopStages({ config, logger, emitter, eventBase, ses
 
 
   // --- Researcher → Planner ---
-  const { plannedTask } = await runPlanningPhases({ config: updatedConfig, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx });
+  const { plannedTask } = await runPlanningPhases({ config: updatedConfig, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx, persistStage, resumeSkip });
 
   // --- Update .gitignore with stack-specific entries based on planner/architect output ---
   const projectDir = updatedConfig.projectDir || process.cwd();
@@ -377,7 +402,7 @@ export async function runPreLoopStages({ config, logger, emitter, eventBase, ses
 
   return { plannedTask, updatedConfig };
 }
-async function runPlanningPhases({ config, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx }) {
+async function runPlanningPhases({ config, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx, persistStage, resumeSkip }) {
   let researchContext = null;
   let plannedTask = task;
 
@@ -386,41 +411,62 @@ async function runPlanningPhases({ config, logger, emitter, eventBase, session,
     ? (await import("../brain-coordinator.js")).processRoleOutput
     : null;
 
+  // KJC-BUG-0058: on resume, rehydrate cross-stage context from session.
+  // researchContext / architectContext / plannedTask live ONLY in memory
+  // after their owning stage runs; without these the post-loop coder
+  // would lose context even though we skipped the upstream LLM call.
+  const bundles = session?.stage_bundles || {};
+
   if (pipelineFlags.researcherEnabled) {
-    const researcherResult = await runResearcherStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
-    researchContext = researcherResult.researchContext;
-    stageResults.researcher = researcherResult.stageResult;
-    if (brainCompress) brainCompress(brainCtx, { roleName: "researcher", output: researcherResult.stageResult, iteration: 0 });
+    if (resumeSkip("researcher")) {
+      researchContext = bundles.researcher?.researchContext || null;
+    } else {
+      const researcherResult = await runResearcherStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
+      researchContext = researcherResult.researchContext;
+      setStageBundle(session, "researcher", { stageResult: researcherResult.stageResult, researchContext });
+      await persistStage("researcher", researcherResult.stageResult);
+      if (brainCompress) brainCompress(brainCtx, { roleName: "researcher", output: researcherResult.stageResult, iteration: 0 });
+    }
   }
 
   // --- Architect (between researcher and planner) ---
   let architectContext = null;
   if (pipelineFlags.architectEnabled) {
-    const architectResult = await runArchitectStage({
-      config, logger, emitter, eventBase, session, coderRole, trackBudget,
-      researchContext,
-      discoverResult: stageResults.discover || null,
-      triageLevel: stageResults.triage?.level || null,
-      askQuestion
-    });
-    architectContext = architectResult.architectContext;
-    stageResults.architect = architectResult.stageResult;
-    if (brainCompress) brainCompress(brainCtx, { roleName: "architect", output: architectResult.stageResult, iteration: 0 });
+    if (resumeSkip("architect")) {
+      architectContext = bundles.architect?.architectContext || null;
+    } else {
+      const architectResult = await runArchitectStage({
+        config, logger, emitter, eventBase, session, coderRole, trackBudget,
+        researchContext,
+        discoverResult: stageResults.discover || null,
+        triageLevel: stageResults.triage?.level || null,
+        askQuestion
+      });
+      architectContext = architectResult.architectContext;
+      setStageBundle(session, "architect", { stageResult: architectResult.stageResult, architectContext });
+      await persistStage("architect", architectResult.stageResult);
+      if (brainCompress) brainCompress(brainCtx, { roleName: "architect", output: architectResult.stageResult, iteration: 0 });
+    }
   }
 
   const triageDecomposition = stageResults.triage?.shouldDecompose ? stageResults.triage.subtasks : null;
   if (pipelineFlags.plannerEnabled) {
-    const plannerRole = resolveRole(config, "planner");
-    const plannerResult = await runPlannerStage({ config, logger, emitter, eventBase, session, plannerRole, researchContext, architectContext, triageDecomposition, trackBudget });
-    plannedTask = plannerResult.plannedTask;
-    stageResults.planner = plannerResult.stageResult;
-    if (brainCompress) brainCompress(brainCtx, { roleName: "planner", output: plannerResult.stageResult, iteration: 0 });
-
-    await tryCiComment({
-      config, session, logger,
-      agent: "Planner",
-      body: `Plan: ${plannerResult.stageResult?.summary || plannedTask}`
-    });
+    if (resumeSkip("planner")) {
+      plannedTask = bundles.planner?.plannedTask || task;
+    } else {
+      const plannerRole = resolveRole(config, "planner");
+      const plannerResult = await runPlannerStage({ config, logger, emitter, eventBase, session, plannerRole, researchContext, architectContext, triageDecomposition, trackBudget });
+      plannedTask = plannerResult.plannedTask;
+      setStageBundle(session, "planner", { stageResult: plannerResult.stageResult, plannedTask });
+      await persistStage("planner", plannerResult.stageResult);
+      if (brainCompress) brainCompress(brainCtx, { roleName: "planner", output: plannerResult.stageResult, iteration: 0 });
+
+      await tryCiComment({
+        config, session, logger,
+        agent: "Planner",
+        body: `Plan: ${plannerResult.stageResult?.summary || plannedTask}`
+      });
+    }
   }
 
   return { plannedTask };

package/src/session/mutators.js

@@ -138,6 +138,37 @@ export function setStatus(session, status) {
   session.status = status;
 }
 
+// --- Stage results (KJC-BUG-0058: resume must skip completed stages) ------
+
+/**
+ * Persist a pre-loop stage result on the session, plus mirror its name in
+ * a flat `stages_completed` array used by `resumeFlow` to know what to skip.
+ *
+ * Both shapes are kept in sync so callers can read either: the rich
+ * `stage_results[name]` for the full payload, or the cheap
+ * `stages_completed` for a fast membership check.
+ */
+export function setStageResult(session, name, result) {
+  if (!session.stage_results) session.stage_results = {};
+  session.stage_results[name] = result;
+  if (!Array.isArray(session.stages_completed)) session.stages_completed = [];
+  if (!session.stages_completed.includes(name)) session.stages_completed.push(name);
+}
+
+/**
+ * Stage *bundles* capture cross-stage context that a stage exports beyond
+ * its stageResult (e.g. researcher → researchContext, architect →
+ * architectContext, planner → plannedTask). Resume needs these to feed
+ * downstream stages without re-running the upstream one. The bundle is
+ * always shaped `{ stageResult, ...extras }` and `setStageBundle` mirrors
+ * the stageResult into `stage_results[name]` so legacy readers keep working.
+ */
+export function setStageBundle(session, name, bundle) {
+  if (!session.stage_bundles) session.stage_bundles = {};
+  session.stage_bundles[name] = bundle;
+  setStageResult(session, name, bundle?.stageResult ?? bundle);
+}
+
 // --- Alternative agent (rate-limit recovery) ------------------------------
 
 export function setAlternativeAgent(session, stage, provider) {