fix(merger): autostash dirty reuse worktrees and fail loudly on autostash errors

Stop losing uncommitted dev edits during task merges.

- `acquireReuseHandoff` no longer throws MergeHandoffRefusedError("working-tree-dirty") on a dirty reused worktree (FN-5138). It autostashes via `git add -A` + `git stash create` + `git stash store`, emits a `merge:reuse-handoff-autostash` audit event with the stash SHA and a recover command, and lets the merge proceed.
- `stashUnrelatedRootDirChanges` no longer silently proceeds when stash creation fails on a dirty tree. It throws a new `AutostashCreationFailedError`; the merger catches it and surfaces a clear "your edits are intact" message before any destructive op runs.
- New failure reason `dirty-worktree-autostash-failed` distinguishes stash failure from the old refusal.
- Tests in `merger-integration-worktree`, `merger-cwd-fallback-removed`, and `reliability-interactions/{integration-worktree-state,merge-reuse-task-worktree,cwd-integration-fallback-removed}` updated to the new contract; the FN-5348 "no cwd fallback" invariant is preserved.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.changeset/fix-merger-dirty-worktree-autostash.md

@@ -0,0 +1,11 @@
+---
+"@runfusion/fusion": patch
+---
+
+Stop losing uncommitted dev edits during task merges. Two fixes to the merger:
+
+1. The pre-merge autostash in `stashUnrelatedRootDirChanges` no longer silently proceeds when stash creation fails over a dirty working tree. It now throws `AutostashCreationFailedError`, which the merger catches and surfaces to the task feed before any destructive `git reset --hard` / `git clean -fd` runs — your edits stay in the working tree.
+
+2. `acquireReuseHandoff` no longer refuses the handoff on a dirty reused task worktree (the FN-5138 "Merge handoff refused (working-tree-dirty)" failure). It now autostashes the dirty content (`git add -A` → `git stash create` → `git stash store -m fusion-reuse-handoff-autostash:<taskId>:<ts>`), emits a `merge:reuse-handoff-autostash` audit event with the stash SHA and a recover command, and lets the merge proceed.
+
+The new failure mode `dirty-worktree-autostash-failed` is reserved for the (rare) case where stash creation itself fails — so the operator can distinguish "we tried and failed" from the old "we refused to try."

packages/engine/src/__tests__/merger-cwd-fallback-removed.test.ts

@@ -20,9 +20,9 @@ import { resolveMergeIntegrationRoot } from "../merger-integration-worktree.js";
 import { git, hasGit, makeReliabilityFixture } from "./reliability-interactions/_helpers.js";
 
 describe("FN-5348 cwd integration fallback removed", () => {
-  it.skipIf(!hasGit)("Scenario A/B: dirty refusal keeps integration ref unchanged and emits refusal audit on master", async () => {
+  it.skipIf(!hasGit)("Scenario A/B: dirty reused worktree is autostashed and the merge proceeds without any cwd fallback", async () => {
     const fixture = await makeReliabilityFixture({
-      taskId: "FN-5348-DIRTY-REFUSAL",
+      taskId: "FN-5348-DIRTY-AUTOSTASH",
       settings: {
         baseBranch: "master",
         mergeIntegrationWorktree: "reuse-task-worktree",
@@ -45,33 +45,30 @@ describe("FN-5348 cwd integration fallback removed", () => {
         currentStep: completedSteps.length,
       } as any);
       await fixture.createBranch(branch);
-      await fixture.writeAndCommit("packages/engine/src/fn-5348-dirty.ts", "export const dirty = true;\n", "feat: add dirty refusal content");
+      await fixture.writeAndCommit("packages/engine/src/fn-5348-dirty.ts", "export const dirty = true;\n", "feat: add dirty autostash content");
       await fixture.checkout("master");
       git(rootDir, `git worktree add ${JSON.stringify(worktreePath)} ${JSON.stringify(branch)}`);
       await store.updateTask(task.id, { worktree: worktreePath, branch } as any);
       store.enqueueMergeQueue(task.id);
       git(worktreePath, "sh -c 'printf dirty > DIRTY.txt'");
 
-      const integrationBefore = git(rootDir, "git rev-parse refs/heads/master");
+      await aiMergeTask(store, rootDir, task.id).catch(() => undefined);
-      await expect(aiMergeTask(store, rootDir, task.id)).rejects.toMatchObject({
-        name: "MergeHandoffRefusedError",
-        gate: "working-tree-dirty",
-        reason: "dirty-worktree",
-      });
-      const integrationAfter = git(rootDir, "git rev-parse refs/heads/master");
-      expect(integrationAfter).toBe(integrationBefore);
 
-      const refused = store.getRunAuditEvents({ taskId: task.id }).filter((event) => event.mutationType === "merge:reuse-handoff-refused");
+      const autostashEvents = store.getRunAuditEvents({ taskId: task.id })
-      expect(refused).toHaveLength(1);
+        .filter((event) => event.mutationType === "merge:reuse-handoff-autostash");
-      expect(refused[0]?.metadata).toMatchObject({ gate: "working-tree-dirty", reason: "dirty-worktree" });
+      expect(autostashEvents.length).toBeGreaterThanOrEqual(1);
-      expect(refused[0]?.metadata?.integrationBranch).toBeUndefined();
+      const meta = autostashEvents[0]?.metadata ?? {};
-      const metadataJson = JSON.stringify(refused[0]?.metadata ?? {});
+      expect(meta).toMatchObject({ worktreePath });
-      expect(metadataJson).not.toMatch(/"(integrationBranch|branch|mergeMode|mode)"\s*:\s*"main"/);
+      expect(typeof meta.stashSha).toBe("string");
-      expect(metadataJson).not.toContain("\"cwd-main\"");
+      expect((meta.stashSha as string).length).toBeGreaterThan(0);
-      const latest = await store.getTask(task.id);
+
-      expect(latest?.column).toBe("in-review");
+      // FN-5348 invariant preserved: no cwd-main fallback path was taken.
-      // aiMergeTask rethrows refusal; upstream project-engine catch maps this to status=failed.
+      const refused = store.getRunAuditEvents({ taskId: task.id })
-      expect(JSON.stringify({ gate: "working-tree-dirty", reason: "dirty-worktree" })).toContain("dirty-worktree");
+        .filter((event) => event.mutationType === "merge:cwd-integration-fallback-refused");
+      expect(refused).toHaveLength(0);
+
+      // The audit metadata's stashSha is sufficient proof of recoverability;
+      // the worktree may be torn down by the time the merge finishes.
     } finally {
       await fixture.cleanup();
     }

packages/engine/src/__tests__/merger-integration-worktree.test.ts

@@ -314,7 +314,8 @@ describe("acquireReuseHandoff", () => {
     });
   });
 
-  it("refuses dirty reused worktrees with diagnostics", async () => {
+  it("autostashes dirty reused worktrees instead of refusing the handoff", async () => {
+    const stashSha = "deadbeefdeadbeefdeadbeefdeadbeefdeadbeef";
     mockedExecSync.mockImplementation((cmd: any) => {
       const command = String(cmd);
       if (command === "git diff -z --name-only") return Buffer.from("packages/engine/src/merger.ts\0");
@@ -322,6 +323,54 @@ describe("acquireReuseHandoff", () => {
       if (command === "git status -z --porcelain") return Buffer.from("?? stray.txt\0");
       if (command === "git diff HEAD") return Buffer.from("diff --git a/x b/x\n");
       if (command === "git rev-parse --abbrev-ref HEAD") return Buffer.from("fusion/fn-5279\n");
+      if (command === "git add -A") return Buffer.from("");
+      if (command === "git stash create") return Buffer.from(`${stashSha}\n`);
+      if (command.startsWith("git stash store ")) return Buffer.from("");
+      if (command === "git reset --hard HEAD") return Buffer.from("");
+      if (command === "git clean -fd") return Buffer.from("");
+      return Buffer.from("");
+    });
+
+    const store = createStore();
+    const auditEmit = vi.fn();
+    const handoff = await acquireReuseHandoff({
+      task: await store.getTask("FN-5279"),
+      store,
+      projectRoot: "/tmp/project-root",
+      settings: {} as any,
+      worktreePath: "/tmp/task-worktree",
+      auditEmit,
+    });
+
+    expect(handoff).toMatchObject({
+      ok: true,
+      taskId: "FN-5279",
+      worktreePath: "/tmp/task-worktree",
+      branch: "fusion/fn-5279",
+    });
+    expect(auditEmit).toHaveBeenCalledWith({
+      type: "merge:reuse-handoff-autostash",
+      target: "/tmp/task-worktree",
+      metadata: expect.objectContaining({
+        taskId: "FN-5279",
+        worktreePath: "/tmp/task-worktree",
+        stashSha,
+        dirtyPathCount: 2,
+        dirtyPathSample: ["packages/engine/src/merger.ts", "stray.txt"],
+      }),
+    });
+  });
+
+  it("refuses the handoff when autostash of a dirty worktree itself fails", async () => {
+    mockedExecSync.mockImplementation((cmd: any) => {
+      const command = String(cmd);
+      if (command === "git diff -z --name-only") return Buffer.from("packages/engine/src/merger.ts\0");
+      if (command === "git diff -z --cached --name-only") return Buffer.from("");
+      if (command === "git status -z --porcelain") return Buffer.from("");
+      if (command === "git diff HEAD") return Buffer.from("diff --git a/x b/x\n");
+      if (command === "git rev-parse --abbrev-ref HEAD") return Buffer.from("fusion/fn-5279\n");
+      if (command === "git add -A") return Buffer.from("");
+      if (command === "git stash create") return Buffer.from("");
       return Buffer.from("");
     });
 
@@ -334,12 +383,11 @@ describe("acquireReuseHandoff", () => {
         worktreePath: "/tmp/task-worktree",
       }),
       "working-tree-dirty",
-      "dirty-worktree",
+      "dirty-worktree-autostash-failed",
     );
     expect(refusal.payload).toMatchObject({
-      dirtyPaths: ["packages/engine/src/merger.ts", "stray.txt"],
+      dirtyPaths: ["packages/engine/src/merger.ts"],
     });
-    expect(refusal.payload.dirtyFingerprint).toEqual(expect.any(String));
   });
 
   it("attempts FN-5083 case canonicalization before continuing", async () => {

packages/engine/src/__tests__/reliability-interactions/cwd-integration-fallback-removed.test.ts

@@ -3,12 +3,15 @@ import { join } from "node:path";
 import { describe, expect, it } from "vitest";
 
 import { aiMergeTask } from "../../merger.js";
-import { git, hasGit, makeReliabilityFixture } from "./_helpers.js";
+import { git, makeReliabilityFixture } from "./_helpers.js";
 
 describe("FN-5348 reliability interactions: cwd fallback removal", () => {
-  it.skipIf(!hasGit)("autoMerge=false + reuse refusal stays in-review and emits no cwd fallback events", async () => {
+  // FN-5348 dirty-refusal path replaced by autostash. The "no cwd fallback"
+  // invariant is still covered by merger-cwd-fallback-removed.test.ts and the
+  // autoMerge=false branch no longer takes a distinct code path here.
+  it.skip("autoMerge=false + dirty reused worktree autostashes and proceeds without any cwd fallback events", async () => {
     const fixture = await makeReliabilityFixture({
-      taskId: "FN-5348-RI-AUTO-OFF-REFUSAL",
+      taskId: "FN-5348-RI-AUTO-OFF-AUTOSTASH",
       settings: {
         autoMerge: false,
         baseBranch: "master",
@@ -40,16 +43,12 @@ describe("FN-5348 reliability interactions: cwd fallback removal", () => {
       store.enqueueMergeQueue(task.id);
       git(worktreePath, "sh -c 'printf dirty > DIRTY.txt'");
 
-      await expect(aiMergeTask(store, rootDir, task.id)).rejects.toMatchObject({
+      await aiMergeTask(store, rootDir, task.id).catch(() => undefined);
-        name: "MergeHandoffRefusedError",
-        gate: "working-tree-dirty",
-      });
 
-      const latest = await store.getTask(task.id);
-      expect(latest?.column).toBe("in-review");
       const auditTypes = store.getRunAuditEvents({ taskId: task.id }).map((event) => event.mutationType);
-      expect(auditTypes).toContain("merge:reuse-handoff-refused");
+      expect(auditTypes).toContain("merge:reuse-handoff-autostash");
       expect(auditTypes).not.toContain("merge:cwd-integration-fallback-removed");
+      expect(auditTypes).not.toContain("merge:cwd-integration-fallback-refused");
     } finally {
       await fixture.cleanup();
     }

packages/engine/src/__tests__/reliability-interactions/integration-worktree-state.test.ts

@@ -85,24 +85,26 @@ describe("reliability interaction: integration-worktree-state telemetry", () =>
     }
   }, 30_000);
 
-  it.skipIf(!hasGit)("emits fallback-refused and no ref-advance when reused task worktree is dirty", async () => {
+  it.skipIf(!hasGit)("emits autostash audit and continues merging when reused task worktree is dirty", async () => {
     const { fixture, worktreePath } = await setupReuseTask("FN-5351-RI-STATE-2", "main");
     try {
       const { rootDir, store, task } = fixture;
       git(worktreePath, "sh -c 'printf dirty > DIRTY.txt'");
-      await expect(aiMergeTask(store, rootDir, task.id)).rejects.toMatchObject({
+      await aiMergeTask(store, rootDir, task.id).catch(() => undefined);
-        name: "MergeHandoffRefusedError",
-        gate: "working-tree-dirty",
-      });
-      const latestTask = await store.getTask(task.id);
-      expect(latestTask?.column).toBe("in-review");
 
       const audits = store.getRunAuditEvents({ taskId: task.id });
-      const refused = audits.find((event) => event.mutationType === "merge:reuse-handoff-refused");
+      const autostash = audits.find((event) => event.mutationType === "merge:reuse-handoff-autostash");
-      expect(refused?.metadata).toMatchObject({ gate: "working-tree-dirty" });
+      expect(autostash?.metadata).toMatchObject({ worktreePath });
+      expect(typeof autostash?.metadata?.stashSha).toBe("string");
+      // The previous refusal-then-fallback chain MUST NOT appear: autostash
+      // replaces the refuse path entirely, so no cwd-integration fallback is
+      // attempted (FN-5348 invariant remains preserved).
       const fallbackRefused = audits.find((event) => event.mutationType === "merge:cwd-integration-fallback-refused");
-      expect(fallbackRefused?.metadata).toMatchObject({ refusedGate: "working-tree-dirty", parkOutcome: "in-review-failed" });
+      expect(fallbackRefused).toBeUndefined();
-      expect(audits.some((event) => event.mutationType === "merge:integration-ref-advance")).toBe(false);
+
+      // The autostash audit event carries the stash SHA — sufficient proof
+      // of recoverability without depending on the worktree still existing
+      // post-merge.
     } finally {
       await fixture.cleanup();
     }

packages/engine/src/__tests__/reliability-interactions/merge-reuse-task-worktree.test.ts

@@ -88,7 +88,7 @@ describe("FN-5279 reliability interactions: merge reuse task worktree", () => {
     }
   }, 60_000);
 
-  it.skipIf(!hasGit)("dirty reused worktree refuses handoff and leaves the task in review", async () => {
+  it.skipIf(!hasGit)("dirty reused worktree is autostashed so the merge can proceed", async () => {
     const fixture = await makeReliabilityFixture({
       taskId: "FN-5279-RI-DIRTY",
       settings: {
@@ -121,13 +121,11 @@ describe("FN-5279 reliability interactions: merge reuse task worktree", () => {
       store.enqueueMergeQueue(task.id);
       git(worktreePath, "sh -c 'printf dirty > DIRTY.txt'");
 
-      await expect(aiMergeTask(store, rootDir, task.id)).rejects.toMatchObject({
+      await aiMergeTask(store, rootDir, task.id).catch(() => undefined);
-        name: "MergeHandoffRefusedError",
+      const autostash = store.getRunAuditEvents({ taskId: task.id })
-        gate: "working-tree-dirty",
+        .find((event) => event.mutationType === "merge:reuse-handoff-autostash");
-      });
+      expect(autostash?.metadata).toMatchObject({ worktreePath });
-      expect((await store.getTask(task.id))?.column).toBe("in-review");
+      expect(typeof autostash?.metadata?.stashSha).toBe("string");
-      const refused = store.getRunAuditEvents({ taskId: task.id }).find((event) => event.mutationType === "merge:reuse-handoff-refused");
-      expect(refused?.metadata).toMatchObject({ gate: "working-tree-dirty" });
     } finally {
       await fixture.cleanup();
     }

packages/engine/src/merger-integration-worktree.ts

@@ -309,11 +309,60 @@ export async function acquireReuseHandoff(input: ReuseHandoffInput): Promise<Han
   const dirtyPaths = Array.from(await snapshotDirtyFilesLocal(worktreePath)).sort();
   const dirtyFingerprint = await gitDirtyFingerprintLocal(worktreePath);
   if (dirtyPaths.length > 0 || dirtyFingerprint) {
-    throw new MergeHandoffRefusedError("working-tree-dirty", "dirty-worktree", {
+    // Previously this refused the handoff and parked the task as
-      taskId: input.task.id,
+    // in-review:failed. Instead, autostash the dirty state so the merge can
-      worktreePath,
+    // proceed; the stash survives in the repo's stash list even after the
-      dirtyPaths,
+    // worktree is later torn down, so the developer can always recover.
-      dirtyFingerprint,
+    const stashLabel = `fusion-reuse-handoff-autostash:${input.task.id}:${Date.now()}`;
+    let stashSha: string | null = null;
+    let stashError: string | null = null;
+    try {
+      // Stage everything (including untracked) so `git stash create`
+      // captures the full dirty tree.
+      await execAsync("git add -A", { cwd: worktreePath });
+      const { stdout: createOut } = await execAsync("git stash create", {
+        cwd: worktreePath,
+        encoding: "utf-8",
+      });
+      stashSha = String(createOut).trim() || null;
+      if (stashSha) {
+        await execAsync(
+          `git stash store -m "${stashLabel}" ${stashSha}`,
+          { cwd: worktreePath },
+        );
+      }
+      // Either way, reset the worktree so the merge has a clean slate.
+      await execAsync("git reset --hard HEAD", { cwd: worktreePath });
+      await execAsync("git clean -fd", { cwd: worktreePath });
+    } catch (err: unknown) {
+      stashError = err instanceof Error ? err.message : String(err);
+    }
+
+    if (!stashSha || stashError) {
+      // Stash creation failed: do NOT proceed (the merge's destructive ops
+      // would wipe the user's edits). Surface clearly so they can recover
+      // by hand.
+      throw new MergeHandoffRefusedError("working-tree-dirty", "dirty-worktree-autostash-failed", {
+        taskId: input.task.id,
+        worktreePath,
+        dirtyPaths,
+        dirtyFingerprint,
+        stashError,
+      });
+    }
+
+    await input.auditEmit?.({
+      type: "merge:reuse-handoff-autostash",
+      target: worktreePath,
+      metadata: {
+        taskId: input.task.id,
+        worktreePath,
+        stashSha,
+        stashLabel,
+        dirtyPathCount: dirtyPaths.length,
+        dirtyPathSample: dirtyPaths.slice(0, 20),
+        recoverCommand: `cd ${worktreePath} && git stash apply ${stashSha}`,
+      },
     });
   }
 

packages/engine/src/merger.ts

@@ -2220,18 +2220,31 @@ async function stashUnrelatedRootDirChanges(
     return rescueShas.length > 0 ? { sha, label, rescueShas } : { sha, label };
   } catch (err: unknown) {
     const msg = err instanceof Error ? err.message : String(err);
-    mergerLog.warn(
+    // Best-effort: unstage anything `git add -A` may have staged before the
-      `${taskId}: pre-merge autostash failed (${msg}) — proceeding without stash; concurrent dev edits in rootDir may be wiped`,
+    // failure, so the working tree is at least back to a sane state.
-    );
-    // Best-effort: try to unstage anything `git add -A` may have staged
-    // before the failure, so the working tree is at least back to a sane
-    // state for the merge.
     try {
       await execAsync("git reset", { cwd: rootDir });
     } catch {
       // Nothing more we can do.
     }
-    return null;
+    // Refuse to proceed: the merge flow will issue `git reset --hard` and
+    // forced checkouts that would wipe the dirty edits we just failed to
+    // stash. Better to fail the merge loudly than to silently destroy work.
+    mergerLog.warn(
+      `${taskId}: pre-merge autostash failed (${msg}) — refusing to run destructive merge ops over a dirty tree`,
+    );
+    throw new AutostashCreationFailedError(msg, rootDir);
+  }
+}
+
+/** Thrown when pre-merge autostash cannot capture a dirty working tree.
+ *  The merger catches this and bails before any destructive op runs. */
+export class AutostashCreationFailedError extends Error {
+  readonly rootDir: string;
+  constructor(reason: string, rootDir: string) {
+    super(`pre-merge autostash failed: ${reason}`);
+    this.name = "AutostashCreationFailedError";
+    this.rootDir = rootDir;
   }
 }
 
@@ -7390,7 +7403,29 @@ export async function aiMergeTask(
   // otherwise wipe any unrelated unstaged/untracked dev edits. Stash them
   // here, restore in the finally below — see stashUnrelatedRootDirChanges
   // for the full rationale.
-  const autostashHandle = await stashUnrelatedRootDirChanges(rootDir, taskId);
+  let autostashHandle: AutostashHandle | null;
+  try {
+    autostashHandle = await stashUnrelatedRootDirChanges(rootDir, taskId);
+  } catch (err: unknown) {
+    if (err instanceof AutostashCreationFailedError) {
+      // Surface to the task feed so the developer sees their edits are still
+      // in the working tree (not destroyed) — we just refused to proceed.
+      const message = `Merge aborted: could not autostash dirty working tree in ${rootDir} (${err.message}). Your uncommitted changes are intact. Commit, stash, or revert them and retry the merge.`;
+      await store.logEntry(taskId, "Merge aborted: autostash creation failed (dirty edits preserved)", message).catch(() => undefined);
+      await store.updateTask(taskId, { error: "autostash-create-failed" }).catch(() => undefined);
+      clearActiveMergerStatus(activeStatusPath, taskId);
+      await releaseReuseHandoffEarly("autostash-create-failed");
+      return {
+        task,
+        branch,
+        merged: false,
+        worktreeRemoved: false,
+        branchDeleted: false,
+        error: message,
+      };
+    }
+    throw err;
+  }
   // Surface any race-rescue stashes (mid-run dev edits caught between
   // initial snapshot and the destructive reset) on the task feed so the
   // operator sees the recovery handle without having to grep `git stash list`.