Files
PaperClipAI/packages/adapter-utils/src/workspace-checkpoint.test.ts
T
DottaandClaude Opus 5.5 c45c7cb854 fix(workspaces): ignore killed-transfer scratch in sparse checkpoints (#19730)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - A native run in a remote sandbox copies its workspace back to the
host after the turn. A sparse checkpoint copies only changed entries,
and the host validates each entry against the sandbox's own manifest
> - Every Daytona transfer stages a `.paperclip-upload-<uuid>` scratch
file at the sandbox workspace root and removes it only in the same
process
> - A control-plane kill during any transfer leaves that scratch behind.
The sparse checkpoint then recorded it as a workspace file, but
providers never export it. Every later copyback and every saved-work
recovery attempt failed with "Workspace checkpoint payload mismatch"
> - The failed recovery then left an accepted Retry showing "Paperclip
is finishing the saved result" with no run
> - This pull request stops recording transfer scratch, makes the
checkpoint export an exact copy, names the failing entry, and gives a
Retry behind a spent recovery an honest reason and a working action
> - The benefit is that a restart during a transfer no longer makes a
task's saved work permanently unrecoverable

## Linked Issues or Issue Description

Refs #19721 (RC-50: the killed copyback owner; this PR covers the
separate RC-50b cause)

**What happened?**

Restart-chaos waves s12/t01 and s14/t05 (Claude on Daytona). After a
SIGKILL of the control plane, the native result was accepted, and then
workspace copyback failed with "Workspace checkpoint payload mismatch".
In s14 all 6 saved-work recovery attempts failed the same way. A Retry
accepted during recovery kept `retryWait.reason =
finalizing_saved_result` after the recovery budget was spent. No run was
created.

**Steps to reproduce**

1. Start a native Claude run in a Daytona sandbox.
2. SIGKILL the control plane while any provider transfer is in flight.
3. Let the run finish its turn and copy its workspace back.

**Expected behavior**

Copyback succeeds. Provider scratch is not workspace content. A Retry
behind a spent recovery names the failure and can start a new recovery.

**Actual behavior**

The s14 sandbox (6f4a0b6a) still held
`/home/daytona/paperclip-workspace/.paperclip-upload-d99fe36b-5acc-437b-96ad-85d5ff9e79c4.tar`.
Its mtime was 03:29:23, the second of the K4 SIGKILL. The Daytona
provider stages transfer scratch at `remoteDir`, which is the lease
`remoteCwd` (the workspace root). The runtime workspace wipe
deliberately preserves `.paperclip-upload-*`. The checkpoint helper
(`workspace-checkpoint.ts`, `checkpointMain`) recorded the stale tar.
The ordinary export excludes `.paperclip-upload*`, so the host received
nothing for that entry. `validateCheckpointPayload` then threw "payload
mismatch" on every attempt. The file-mode hypothesis was disproved: the
sandbox umask is 022 and the scratch is mode 644.

## What Changed

- `packages/adapter-utils/src/exclude-patterns.ts`: new
`isSandboxTransferScratchPath`, which matches a root-level
`.paperclip-upload-*` entry. This agrees with the wipe's preserve rule
in `buildWorkspaceTarExtractCommand`.
- `packages/adapter-utils/src/workspace-checkpoint.ts`: the sandbox
checkpoint helper skips that scratch. A mismatch error now names the
entry and its recorded, baseline and received forms (kind, octal mode,
short hash; never contents).
- `packages/adapter-utils/src/sandbox-managed-runtime.ts`: the
checkpoint export requests `preserveSourceEntries: true`. The checkpoint
output is an exact copy that the host validates entry by entry. An
ordinary export applies the host umask and drops `._*` and
`.paperclip-upload*` names at any depth. That gives the same permanent
mismatch for a legitimate user file such as `._summary`,
`notes/.paperclip-upload-plan.md`, or a `0775` script.
- `server/src/services/explicit-native-continuation.ts` and
`task-interruption-presentation.ts`: when the run's saved-work recovery
is exhausted, a waiting Retry reports `saved_work_recovery_failed` with
Retry available, instead of `finalizing_saved_result`. The existing task
Retry path starts one new recovery generation.
- `doc/execution-semantics.md`: a short additive note.

## Verification

- `packages/adapter-utils/src/workspace-checkpoint-capture.test.ts` uses
the real checkpoint helper and an export that behaves like Daytona's
ordinary export:
- Before the fix: `Workspace checkpoint payload mismatch at
".paperclip-upload-d99fe36b-….tar": sandbox recorded file mode=644 …;
baseline absent; received absent`. That is the wave error.
  - After the fix: the scratch is not recorded, and validation passes.
- A second case shows that reserved-looking user names and `0775` modes
need the exact export.
- `packages/adapter-utils/src/workspace-checkpoint.test.ts` is an
end-to-end `prepareSandboxManagedRuntime` restore through a provider
model. In that model, ordinary exports drop reserved names and apply the
umask, and preserve exports copy exactly. Two red checks:
- Without `preserveSourceEntries`, it fails with the mismatch on
`._summary`.
- Without the scratch skip, the stale tar leaks into the host workspace.
- Green: the agent file, the `._summary` file and the `0775` mode
arrive, and no scratch arrives.
- `server/src/services/workspace-recovery-preparation-gates.test.ts`,
"names a spent saved-work recovery behind an accepted Retry, and the
next Retry finishes it" (s14 shape):
  - Red: `finalizing_saved_result` after 6 failed attempts.
- Green: `saved_work_recovery_failed` with `canRetry: true`. The next
Retry starts a new generation, recovery completes, and the Retry
continues in exactly one new run.
- Suites: adapter-utils checkpoint, capture, sandbox-managed-runtime,
seed-reuse and managed-seed (168 passed). Server preparation-gates (47),
explicit-native-continuation and task-interruption-presentation (455),
workspace-recovery admission and retry-copyback, and
native-workspace-finalizer. `tsc --noEmit` is clean for adapter-utils
and server.
- Live evidence: I started the stopped s14 sandbox read-only, listed its
workspace, and stopped it again. No Daytona replay was run. The runner
content ID of this checkout matches no built image.

## Risks

- A user file named `.paperclip-upload-*` directly at the workspace root
is no longer captured by a sparse checkpoint. Providers already drop
that name from every ordinary export, and the runtime already reserves
it at the root.
- The checkpoint export now uses the provider's exact-copy mode. A
workspace file with setuid, setgid or sticky bits is refused by that
mode, as it is for the existing exact-copy transfers. The ordinary
export would have stripped those bits and failed validation anyway.
- The stale scratch tar stays in the sandbox. It is ignored, not
deleted, because a concurrent transfer can own a file with the same
prefix.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, through Claude Code
with tool use (shell, file edits, test runs, embedded Postgres, a
read-only Daytona sandbox listing).

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [ ] I will address all Greptile and reviewer comments before
requesting merge

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-06 22:46:16 -06:00

507 lines
35 KiB
TypeScript

import fs from "node:fs/promises";
import os from "node:os";
import path from "node:path";
import { execFile } from "node:child_process";
import { promisify } from "node:util";
import { afterEach, describe, expect, it, vi } from "vitest";
import { captureDirectorySnapshot, disposeDirectorySnapshot, mergeDirectoryWithBaseline } from "./workspace-restore-merge.js";
import { writeCheckpointBaseline, workspaceCheckpointScript, readCheckpointSnapshot, validateCheckpointPayload } from "./workspace-checkpoint.js";
import { publishWorkspaceSeedGeneration, readWorkspaceSeedGeneration, workspaceSeedGeneration } from "./workspace-seed-cache.js";
import { WorkspaceManifestWriter } from "./workspace-manifest.js";
const execute = promisify(execFile);
const roots: string[] = [];
afterEach(async () => { vi.unstubAllEnvs(); await Promise.all(roots.splice(0).map((root) => fs.rm(root, { recursive: true, force: true }))); });
async function fixture() {
const temp = await fs.mkdtemp(path.join(os.tmpdir(), "checkpoint-test-")); roots.push(temp); vi.stubEnv("PAPERCLIP_HOME", path.join(temp, "home"));
const root = path.join(temp, "root"); await fs.mkdir(root);
await fs.writeFile(path.join(root, "unchanged"), "retained");
await fs.writeFile(path.join(root, "deleted"), "delete me");
await fs.mkdir(path.join(root, "cache")); await fs.writeFile(path.join(root, "cache", "ignored"), "private");
const baseline = await captureDirectorySnapshot(root, { diskBacked: true, exclude: ["cache"] });
const manifest = path.join(temp, "baseline.sqlite"); await writeCheckpointBaseline(baseline, manifest);
const script = path.join(temp, "checkpoint.mjs"); await fs.writeFile(script, workspaceCheckpointScript());
const capture = async () => {
const output = path.join(temp, "output");
await execute(process.execPath, [script, root, manifest, output, JSON.stringify(baseline.exclude)]);
const snapshot = readCheckpointSnapshot(path.join(output, "manifest.sqlite"), baseline);
const payload = path.join(output, "payload");
const metrics = await validateCheckpointPayload({ baseline, snapshot, payload });
return { snapshot, payload, metrics, receipt: JSON.parse(await fs.readFile(path.join(output, "receipt.json"), "utf8")) };
};
return { temp, root, baseline, capture };
}
describe("workspace sparse checkpoint", () => {
it.each((["file", "symlink", "missing"] as const).flatMap(attack =>
(["a/kept", "a/nested/kept"] as const).flatMap(child =>
(["manifest", "memory"] as const).map(storage => ({ attack, child, storage })))))(
"rejects a returned $attack ancestor of $child ($storage) before any host mutation", async ({ attack, child, storage }) => {
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
await fs.mkdir(path.dirname(path.join(f.root, child)), { recursive: true });
await fs.writeFile(path.join(f.root, child), "retained child");
const baseline = await captureDirectorySnapshot(f.root, { diskBacked: true, exclude: ["cache"] });
const payload = path.join(f.temp, "forged-payload"); await fs.mkdir(payload);
await fs.writeFile(path.join(payload, "unchanged"), "otherwise valid edit");
if (attack === "file") await fs.writeFile(path.join(payload, "a"), "replace directory");
if (attack === "symlink") await fs.symlink("unchanged", path.join(payload, "a"));
const captured = await captureDirectorySnapshot(payload);
const entries = new Map(baseline.entries);
entries.delete("deleted"); entries.delete("a");
for (const [relative, entry] of captured.entries) entries.set(relative, entry);
const manifest = path.join(f.temp, "forged.sqlite");
await writeCheckpointBaseline({ ...baseline, entries }, manifest);
const evidence = await fs.readFile(manifest);
const snapshot = storage === "manifest" ? readCheckpointSnapshot(manifest, baseline)
: { ...baseline, entries: new Map([...entries].reverse()) };
let failure: unknown;
try {
await validateCheckpointPayload({ baseline, snapshot, payload });
await mergeDirectoryWithBaseline({ baseline, sourceDir: payload, targetDir: f.root, snapshots: { source: snapshot } });
} catch (error) { failure = error; }
expect(await fs.readFile(path.join(f.root, child), "utf8")).toBe("retained child");
expect(await fs.readFile(path.join(f.root, "unchanged"), "utf8")).toBe("retained");
expect(await fs.readFile(path.join(f.root, "deleted"), "utf8")).toBe("delete me");
expect(failure).toBeInstanceOf(Error);
expect(String(failure)).toContain("Invalid workspace checkpoint parent");
expect(await fs.readFile(manifest)).toEqual(evidence);
const replay = readCheckpointSnapshot(manifest, baseline);
await expect(validateCheckpointPayload({ baseline, snapshot: replay, payload })).rejects.toThrow("Invalid workspace checkpoint parent");
await disposeDirectorySnapshot(replay); await disposeDirectorySnapshot(snapshot); await disposeDirectorySnapshot(baseline);
});
it.each(["file_to_directory", "directory_to_file"] as const)("restores a coherent %s checkpoint", async transition => {
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
if (transition === "file_to_directory") await fs.writeFile(path.join(f.root, "a"), "old file");
else {
await fs.mkdir(path.join(f.root, "a")); await fs.writeFile(path.join(f.root, "a", "old"), "old child");
}
const baseline = await captureDirectorySnapshot(f.root, { diskBacked: true, exclude: ["cache"] });
const host = path.join(f.temp, "host"); await fs.cp(f.root, host, { recursive: true });
await fs.rm(path.join(f.root, "a"), { recursive: true });
if (transition === "file_to_directory") {
await fs.mkdir(path.join(f.root, "a", "nested"), { recursive: true });
await fs.writeFile(path.join(f.root, "a", "nested", "new"), "new child");
} else await fs.writeFile(path.join(f.root, "a"), "new file");
const manifest = path.join(f.temp, "transition-baseline.sqlite"); await writeCheckpointBaseline(baseline, manifest);
const output = path.join(f.temp, "transition-output");
await execute(process.execPath, [path.join(f.temp, "checkpoint.mjs"), f.root, manifest, output, JSON.stringify(baseline.exclude)]);
const snapshot = readCheckpointSnapshot(path.join(output, "manifest.sqlite"), baseline), payload = path.join(output, "payload");
await validateCheckpointPayload({ baseline, snapshot, payload });
await mergeDirectoryWithBaseline({ baseline, sourceDir: payload, targetDir: host, snapshots: { source: snapshot } });
expect(await fs.readFile(path.join(host, transition === "file_to_directory" ? "a/nested/new" : "a"), "utf8"))
.toBe(transition === "file_to_directory" ? "new child" : "new file");
expect(await fs.readFile(path.join(host, "unchanged"), "utf8")).toBe("retained");
await disposeDirectorySnapshot(snapshot); await disposeDirectorySnapshot(baseline);
});
it("enumerates unchanged files but transfers zero workspace payload", async () => {
const f = await fixture(); const result = await f.capture();
expect(result.receipt).toMatchObject({ files: 0, bytes: 0, scanned: 2 });
expect(result.metrics).toEqual({ mode: "sparse", scannedEntries: 2, changedFiles: 0, payloadBytes: 0 });
expect(await fs.readdir(result.payload)).toEqual([]);
expect([...result.snapshot.entries]).toEqual([...f.baseline.entries]);
await disposeDirectorySnapshot(result.snapshot); await disposeDirectorySnapshot(f.baseline);
});
it("merges one changed file, deletion, unusual names and a safe link without copying unchanged files", async () => {
const f = await fixture(); const host = path.join(f.temp, "host"); await fs.cp(f.root, host, { recursive: true });
const name = "new\nfile"; await fs.writeFile(path.join(f.root, name), "changed");
await fs.unlink(path.join(f.root, "deleted")); await fs.symlink(name, path.join(f.root, "link"));
const result = await f.capture(); expect(result.receipt.files).toBe(1); expect(result.receipt.bytes).toBe(7);
await mergeDirectoryWithBaseline({ baseline: f.baseline, sourceDir: result.payload, targetDir: host, snapshots: { source: result.snapshot } });
expect(await fs.readFile(path.join(host, name), "utf8")).toBe("changed");
expect(await fs.readFile(path.join(host, "unchanged"), "utf8")).toBe("retained");
expect(await fs.readlink(path.join(host, "link"))).toBe(name);
await expect(fs.stat(path.join(host, "deleted"))).rejects.toMatchObject({ code: "ENOENT" });
expect(await fs.readFile(path.join(host, "cache", "ignored"), "utf8")).toBe("private");
await disposeDirectorySnapshot(f.baseline);
});
it.each(["absolute", "escaping"])("omits an %s link while restoring ordinary changes and safe links", async (kind) => {
const f = await fixture();
const host = path.join(f.temp, "host"); await fs.cp(f.root, host, { recursive: true });
const outside = path.join(f.temp, "secret"); await fs.writeFile(outside, "outside bytes");
await fs.symlink(kind === "absolute" ? outside : "../secret", path.join(f.root, "unsafe"));
await fs.writeFile(path.join(f.root, "unchanged"), "edited");
await fs.writeFile(path.join(f.root, "new"), "new output");
await fs.unlink(path.join(f.root, "deleted"));
await fs.symlink("new", path.join(f.root, "safe"));
const result = await f.capture();
expect(result.snapshot.entries.has("unsafe")).toBe(false);
await expect(fs.lstat(path.join(result.payload, "unsafe"))).rejects.toMatchObject({ code: "ENOENT" });
await mergeDirectoryWithBaseline({ baseline: f.baseline, sourceDir: result.payload, targetDir: host, snapshots: { source: result.snapshot } });
expect(await fs.readFile(path.join(host, "unchanged"), "utf8")).toBe("edited");
expect(await fs.readFile(path.join(host, "new"), "utf8")).toBe("new output");
expect(await fs.readlink(path.join(host, "safe"))).toBe("new");
await expect(fs.lstat(path.join(host, "unsafe"))).rejects.toMatchObject({ code: "ENOENT" });
await expect(fs.lstat(path.join(host, "deleted"))).rejects.toMatchObject({ code: "ENOENT" });
expect(await fs.readFile(outside, "utf8")).toBe("outside bytes");
await disposeDirectorySnapshot(result.snapshot); await disposeDirectorySnapshot(f.baseline);
});
it("rejects an unsafe link forged into a checkpoint manifest before merge", async () => {
const f = await fixture(); const result = await f.capture();
const entries = new Map(result.snapshot.entries);
entries.set("unsafe", { kind: "symlink", target: "../secret" });
await expect(validateCheckpointPayload({ baseline: f.baseline, snapshot: { ...result.snapshot, entries }, payload: result.payload }))
.rejects.toMatchObject({ code: "WORKSPACE_RESTORE_UNSAFE_ARCHIVE" });
await disposeDirectorySnapshot(result.snapshot); await disposeDirectorySnapshot(f.baseline);
});
it("replays a checkpoint after interruption and preserves a concurrently edited deletion", async () => {
const f = await fixture(); const host = path.join(f.temp, "host"); await fs.cp(f.root, host, { recursive: true });
await fs.writeFile(path.join(f.root, "new"), "remote"); await fs.unlink(path.join(f.root, "deleted"));
await fs.writeFile(path.join(host, "deleted"), "host edit");
const result = await f.capture();
await mergeDirectoryWithBaseline({ baseline: f.baseline, sourceDir: result.payload, targetDir: host, snapshots: { source: result.snapshot } });
const replay = readCheckpointSnapshot(path.join(f.temp, "output", "manifest.sqlite"), f.baseline);
await validateCheckpointPayload({ baseline: f.baseline, snapshot: replay, payload: result.payload });
await mergeDirectoryWithBaseline({ baseline: f.baseline, sourceDir: result.payload, targetDir: host, snapshots: { source: replay } });
expect(await fs.readFile(path.join(host, "deleted"), "utf8")).toBe("host edit");
expect(await fs.readFile(path.join(host, "new"), "utf8")).toBe("remote");
await disposeDirectorySnapshot(f.baseline);
});
it("rejects bytes changed after capture before applying the manifest", async () => {
const f = await fixture(); await fs.writeFile(path.join(f.root, "new"), "remote");
const result = await f.capture(); await fs.writeFile(path.join(result.payload, "new"), "tampered");
await expect(validateCheckpointPayload({ baseline: f.baseline, snapshot: result.snapshot, payload: result.payload })).rejects.toThrow("payload mismatch");
await disposeDirectorySnapshot(result.snapshot); await disposeDirectorySnapshot(f.baseline);
});
it.each(["array", "manifest"] as const)("rejects ignored entries and ancestor replacements before merge with %s ignore paths", async storage => {
for (const attack of ["matching_file", "matching_descendant", "payload_only", "ancestor_file", "ancestor_link"] as const) {
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
await fs.writeFile(path.join(f.root, ".env"), "host secret");
await fs.mkdir(path.join(f.root, "private")); await fs.writeFile(path.join(f.root, "private", "token"), "private token");
await fs.mkdir(path.join(f.root, "config")); await fs.writeFile(path.join(f.root, "config", ".env"), "nested secret");
const paths = [".env", "private", "config/.env"];
const writer = new WorkspaceManifestWriter(path.join(f.temp, "ignored.sqlite"));
for (const relative of paths) writer.add("ignored", relative);
const ignoredPaths = storage === "manifest" ? writer.paths("ignored") : paths;
writer.close();
const baseline = await captureDirectorySnapshot(f.root, { exclude: ["cache"], ignoredPaths, diskBacked: true });
const payload = path.join(f.temp, "forged-payload"); await fs.mkdir(payload);
await fs.writeFile(path.join(payload, "unchanged"), "otherwise valid edit");
if (attack === "matching_file" || attack === "payload_only") await fs.writeFile(path.join(payload, ".env"), "injected secret");
if (attack === "matching_descendant") {
await fs.mkdir(path.join(payload, "private")); await fs.writeFile(path.join(payload, "private", "token"), "injected token");
}
if (attack === "ancestor_file") await fs.writeFile(path.join(payload, "config"), "replace directory");
if (attack === "ancestor_link") await fs.symlink("unchanged", path.join(payload, "config"));
const captured = await captureDirectorySnapshot(payload);
const entries = new Map(baseline.entries);
entries.delete("deleted");
for (const [relative, entry] of captured.entries) if (!(attack === "payload_only" && relative === ".env")) entries.set(relative, entry);
if (attack === "matching_descendant") entries.delete("private");
const manifestPath = path.join(f.temp, "forged.sqlite");
await writeCheckpointBaseline({ ...baseline, entries }, manifestPath);
const snapshot = readCheckpointSnapshot(manifestPath, baseline);
try {
await expect((async () => {
await validateCheckpointPayload({ baseline, snapshot, payload });
await mergeDirectoryWithBaseline({ baseline, sourceDir: payload, targetDir: f.root, snapshots: { source: snapshot } });
})()).rejects.toThrow(attack.startsWith("ancestor") ? "ignored path ancestor" : "Excluded workspace checkpoint entry");
expect(await fs.readFile(path.join(f.root, "unchanged"), "utf8")).toBe("retained");
expect(await fs.readFile(path.join(f.root, "deleted"), "utf8")).toBe("delete me");
expect(await fs.readFile(path.join(f.root, ".env"), "utf8")).toBe("host secret");
expect(await fs.readFile(path.join(f.root, "private", "token"), "utf8")).toBe("private token");
expect(await fs.readFile(path.join(f.root, "config", ".env"), "utf8")).toBe("nested secret");
} finally { await disposeDirectorySnapshot(snapshot); await disposeDirectorySnapshot(baseline); }
}
});
it("allows directory tombstones while retaining ignored host children", async () => {
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
await fs.mkdir(path.join(f.root, "config"));
await fs.writeFile(path.join(f.root, "config", ".env"), "host only");
await fs.writeFile(path.join(f.root, "config", "tracked"), "delete tracked child");
const baseline = await captureDirectorySnapshot(f.root, { exclude: ["cache"], ignoredPaths: ["config/.env"], diskBacked: true });
const payload = path.join(f.temp, "payload"); await fs.mkdir(payload);
await fs.writeFile(path.join(payload, "unchanged"), "valid edit");
const captured = await captureDirectorySnapshot(payload);
const snapshot = { ...baseline, entries: new Map(captured.entries) };
await validateCheckpointPayload({ baseline, snapshot, payload });
await mergeDirectoryWithBaseline({ baseline, sourceDir: payload, targetDir: f.root, snapshots: { source: snapshot } });
expect(await fs.readFile(path.join(f.root, "config", ".env"), "utf8")).toBe("host only");
expect(await fs.readFile(path.join(f.root, "unchanged"), "utf8")).toBe("valid edit");
await expect(fs.stat(path.join(f.root, "config", "tracked"))).rejects.toMatchObject({ code: "ENOENT" });
await expect(fs.stat(path.join(f.root, "deleted"))).rejects.toMatchObject({ code: "ENOENT" });
await disposeDirectorySnapshot(baseline);
});
it("reuses durable seed generations and rejects corruption after run cleanup", async () => {
const f = await fixture(); const cache = path.join(f.temp, "cache-seeds");
const archive = path.join(f.temp, "seed.tar"); await fs.writeFile(archive, "archive");
const generation = workspaceSeedGeneration(f.baseline, null);
await publishWorkspaceSeedGeneration(cache, generation, { workspaceArchivePath: archive });
await fs.unlink(archive);
const cached = await readWorkspaceSeedGeneration(cache, generation); expect(cached).not.toBeNull();
expect(await fs.readFile(cached!.workspaceArchivePath, "utf8")).toBe("archive");
await fs.writeFile(cached!.workspaceArchivePath, "corrupted");
expect(await readWorkspaceSeedGeneration(cache, generation)).toBeNull();
await disposeDirectorySnapshot(f.baseline);
});
});
it("uses sparse native transfer end-to-end, including replay and unsupported-runtime fallback", async () => {
const { prepareSandboxManagedRuntime } = await import("./sandbox-managed-runtime.js");
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
const remote = path.join(f.temp, "remote");
const cache = path.join(f.temp, "seeds");
let supportsCheckpoint = true;
let payloadBytes = -1;
let mutateDuringPack = false;
let failVerification = false, verificationCalls = 0;
const run = async (command: string) => {
if (!supportsCheckpoint && command.startsWith("node --input-type=module")) throw new Error("node unavailable");
await execute("sh", ["-c", command]);
};
const sync = async (operations: import("./sandbox-managed-runtime.js").SandboxSyncOperation[]) => {
for (const operation of operations) {
for (const file of operation.files) {
await fs.mkdir(path.dirname(file.targetPath), { recursive: true });
await fs.cp(file.sourcePath, file.targetPath, { recursive: true, verbatimSymlinks: true,
filter: (source) => !file.exclude?.some((exclude) => path.relative(file.sourcePath, source) === exclude) });
if (file.sourcePath.endsWith("/output")) {
payloadBytes = JSON.parse(await fs.readFile(path.join(file.targetPath, "receipt.json"), "utf8")).bytes;
}
}
for (const command of operation.postUploadCommands ?? []) await run(command.command);
}
return { operations: operations.map((op) => ({ operationId: op.operationId, filesTransferred: 0, bytesTransferred: 0 })) };
};
const client = { makeDir: (dir: string) => fs.mkdir(dir, { recursive: true }).then(() => {}),
writeFile: (file: string, data: ArrayBuffer) => fs.writeFile(file, Buffer.from(data)), readFile: (file: string) => fs.readFile(file),
listFiles: (dir: string) => fs.readdir(dir), remove: (file: string) => fs.rm(file, { force: true, recursive: true }), run,
syncIn: sync, syncOut: sync };
const prepare = (overrides: Partial<Parameters<typeof prepareSandboxManagedRuntime>[0]> = {}) => prepareSandboxManagedRuntime({ spec: { provider: "test", sandboxId: "test", remoteCwd: remote, timeoutMs: 30000, apiKey: null, transport: "sandbox" },
adapterKey: "test", client, workspaceLocalDir: f.root, workspaceCheckpoint: true,
runtimeSpan: async (name, work) => {
if (mutateDuringPack && name === "pack") await fs.writeFile(path.join(f.root, "unchanged"), "concurrent change");
if (name === "seed.verify") {
verificationCalls++;
const result = await work();
if (failVerification) throw Object.assign(new Error("verification disk full"), { code: "ENOSPC" });
return result;
}
return work();
},
workspaceSeedCacheDirectory: cache, workspaceDurableSeed: { workspaceArchivePath: path.join(f.temp, "run-seed.tar") }, ...overrides });
const first = await prepare(); await first.restoreWorkspace(); expect(payloadBytes).toBe(0);
const generation = (await fs.readdir(cache))[0]!;
await fs.writeFile(path.join(cache, generation, "workspace.tar"), "corrupt cache");
const second = await prepare();
await fs.writeFile(path.join(remote, "new"), "changed"); await fs.unlink(path.join(remote, "deleted"));
const outside = path.join(f.temp, "outside"); await fs.writeFile(outside, "private");
await fs.symlink(outside, path.join(remote, "absolute-link"));
await fs.symlink("../outside", path.join(remote, "escaping-link"));
await fs.symlink("new", path.join(remote, "safe-link"));
await second.restoreWorkspace(); expect(payloadBytes).toBe(7);
expect(await fs.readFile(path.join(f.root, "new"), "utf8")).toBe("changed");
expect(await fs.readlink(path.join(f.root, "safe-link"))).toBe("new");
for (const name of ["absolute-link", "escaping-link"]) {
await expect(fs.lstat(path.join(f.root, name))).rejects.toMatchObject({ code: "ENOENT" });
}
expect(await fs.readFile(outside, "utf8")).toBe("private");
await expect(fs.stat(path.join(f.root, "deleted"))).rejects.toMatchObject({ code: "ENOENT" });
supportsCheckpoint = false;
const third = await prepare(); await fs.writeFile(path.join(remote, "new"), "fallback");
await third.restoreWorkspace(); expect(await fs.readFile(path.join(f.root, "new"), "utf8")).toBe("fallback");
const existingGenerations = await fs.readdir(cache);
mutateDuringPack = true;
const raced = await prepare();
expect(await fs.readdir(cache)).toEqual(existingGenerations);
expect((await fs.stat(path.join(f.temp, "run-seed.tar"))).size).toBeGreaterThan(0);
await raced.cleanupWorkspaceSnapshot();
mutateDuringPack = false; failVerification = true;
await fs.writeFile(path.join(f.root, "new"), "verification recovery");
const verificationFailed = await prepare();
expect(await fs.readdir(cache)).toEqual(existingGenerations);
const failedSnapshot = verificationFailed.workspaceSyncSnapshot!;
await fs.writeFile(path.join(f.root, "new"), "later host bytes");
await fs.rm(remote, { recursive: true, force: true });
const verificationRecovery = await prepare({ workspaceInboundMode: "durable_seed", workspaceBaseline: failedSnapshot.baseline,
workspaceGitSnapshot: failedSnapshot.gitSnapshot, workspaceRepositories: failedSnapshot.repositories });
expect(await fs.readFile(path.join(remote, "new"), "utf8")).toBe("verification recovery");
await verificationRecovery.cleanupWorkspaceSnapshot();
failVerification = false;
mutateDuringPack = false;
for (let count = (await fs.readdir(cache)).length; count < 16; count++) {
await fs.mkdir(path.join(cache, `.pending-retained-${count}`));
}
const saturated = await fs.readdir(cache);
await fs.writeFile(path.join(f.root, "new"), "quota recovery");
const verificationCallsBeforeSaturation = verificationCalls;
const uncached = await prepare();
expect(verificationCalls).toBe(verificationCallsBeforeSaturation);
expect(await fs.readdir(cache)).toEqual(saturated);
const snapshot = uncached.workspaceSyncSnapshot!;
await fs.writeFile(path.join(f.root, "new"), "later host edit");
await fs.rm(remote, { recursive: true, force: true });
const recovered = await prepare({ workspaceInboundMode: "durable_seed", workspaceBaseline: snapshot.baseline,
workspaceGitSnapshot: snapshot.gitSnapshot, workspaceRepositories: snapshot.repositories });
expect(await fs.readFile(path.join(remote, "new"), "utf8")).toBe("quota recovery");
expect(await fs.readFile(path.join(f.root, "new"), "utf8")).toBe("later host edit");
expect(await fs.readdir(cache)).toEqual(saturated);
await recovered.cleanupWorkspaceSnapshot();
}, 30000);
async function exportFailureFixture(failure: "low_disk" | "full_root" | "full_special" | "digest" | "path" | "symlink") {
const { prepareSandboxManagedRuntime } = await import("./sandbox-managed-runtime.js");
const { DatabaseSync } = await import("node:sqlite");
const f = await fixture();
await disposeDirectorySnapshot(f.baseline);
const remote = path.join(f.temp, "remote"), outside = path.join(f.temp, "outside");
await fs.writeFile(outside, "private");
const metrics: unknown[] = [];
const counts = { captures: 0, fullExports: 0, capacityFailures: 0 };
const run = async (command: string) => {
const capture = command.includes("checkpoint.mjs");
if (capture) counts.captures++;
try { await execute("sh", ["-c", command]); }
catch (error) {
if (capture && ["low_disk", "full_root", "full_special"].includes(failure)) {
expect(String(error)).toContain("Workspace checkpoint storage is below its free-space reserve");
counts.capacityFailures++;
await expect(fs.stat(path.join(f.root, "new"))).rejects.toMatchObject({ code: "ENOENT" });
}
throw error;
}
};
const sync = async (operations: import("./sandbox-managed-runtime.js").SandboxSyncOperation[]) => {
for (const operation of operations) {
for (const file of operation.files) {
if (file.sourcePath === remote) {
counts.fullExports++;
if (failure === "full_root") {
await fs.rm(file.targetPath, { recursive: true, force: true });
await fs.symlink(remote, file.targetPath);
continue;
}
}
await fs.mkdir(path.dirname(file.targetPath), { recursive: true });
await fs.cp(file.sourcePath, file.targetPath, { recursive: true, verbatimSymlinks: true,
filter: source => !file.exclude?.some(exclude => path.relative(file.sourcePath, source) === exclude) });
if (file.sourcePath === remote && failure === "full_special") await execute("mkfifo", [path.join(file.targetPath, "unsafe-pipe")]);
if (["low_disk", "full_root", "full_special"].includes(failure) && file.targetPath.endsWith("/checkpoint.mjs")) {
const script = await fs.readFile(file.targetPath, "utf8");
await fs.writeFile(file.targetPath, script.replace("const __name", "fs.statfs = async () => ({ bavail: 128n * 1024n, bsize: 1024n });\nconst __name"));
}
if (file.sourcePath.endsWith("/output")) {
if (failure === "digest") await fs.writeFile(path.join(file.targetPath, "payload", "new"), "forged");
else if (failure === "path" || failure === "symlink") {
const manifest = new DatabaseSync(path.join(file.targetPath, "manifest.sqlite"));
try {
manifest.prepare("INSERT INTO records VALUES('baseline',?,?)").run(
failure === "path" ? "../escape" : "escape",
JSON.stringify({ kind: "symlink", target: outside }),
);
} finally { manifest.close(); }
}
}
}
for (const command of operation.postUploadCommands ?? []) await run(command.command);
}
return { operations: operations.map(op => ({ operationId: op.operationId, filesTransferred: 0, bytesTransferred: 0 })) };
};
const client = { makeDir: (dir: string) => fs.mkdir(dir, { recursive: true }).then(() => {}),
writeFile: (file: string, data: ArrayBuffer) => fs.writeFile(file, Buffer.from(data)), readFile: (file: string) => fs.readFile(file),
listFiles: (dir: string) => fs.readdir(dir), remove: (file: string) => fs.rm(file, { force: true, recursive: true }), run,
syncIn: sync, syncOut: sync };
const prepare = (overrides: Partial<Parameters<typeof prepareSandboxManagedRuntime>[0]> = {}) => prepareSandboxManagedRuntime({
spec: { provider: "test", sandboxId: "test", remoteCwd: remote, timeoutMs: 30000, apiKey: null, transport: "sandbox" },
adapterKey: "test", client, workspaceLocalDir: f.root, workspaceCheckpoint: true,
onWorkspaceCheckpoint: async metric => { metrics.push(metric); }, ...overrides,
});
const prepared = await prepare();
await fs.writeFile(path.join(remote, "new"), "tiny export");
await fs.unlink(path.join(remote, "deleted"));
return { ...f, remote, outside, prepared, prepare, metrics, counts };
}
it.each([false, true])("falls back once to a full tiny export below the sparse disk reserve (recovery=%s)", async recovery => {
const f = await exportFailureFixture("low_disk");
const snapshot = f.prepared.workspaceSyncSnapshot!;
const runtime = recovery ? await f.prepare({ workspaceInboundMode: "adopt_remote", workspaceBaseline: snapshot.baseline,
workspaceGitSnapshot: snapshot.gitSnapshot, workspaceRepositories: snapshot.repositories }) : f.prepared;
await fs.writeFile(path.join(f.root, "host-only"), "concurrent host edit");
await fs.symlink(f.outside, path.join(f.remote, "unsafe-link"));
await fs.symlink("../outside", path.join(f.remote, "escaping-link"));
await fs.symlink("new", path.join(f.remote, "safe-link"));
await runtime.restoreWorkspace();
expect(f.counts).toEqual({ captures: 1, fullExports: 1, capacityFailures: 1 });
expect(f.metrics).toEqual([{ mode: "full", scannedEntries: null, changedFiles: null, payloadBytes: null }]);
expect(await fs.readFile(path.join(f.root, "new"), "utf8")).toBe("tiny export");
expect(await fs.readFile(path.join(f.root, "host-only"), "utf8")).toBe("concurrent host edit");
await expect(fs.stat(path.join(f.root, "deleted"))).rejects.toMatchObject({ code: "ENOENT" });
expect(await fs.readFile(f.outside, "utf8")).toBe("private");
for (const name of ["unsafe-link", "escaping-link"]) {
await expect(fs.lstat(path.join(f.root, name))).rejects.toMatchObject({ code: "ENOENT" });
}
expect(await fs.readlink(path.join(f.root, "safe-link"))).toBe("new");
await runtime.cleanupWorkspaceSnapshot();
if (recovery) await f.prepared.cleanupWorkspaceSnapshot();
}, 30000);
it.each(["digest", "path", "symlink"] as const)("never falls back after returned sparse %s validation fails", async attack => {
const f = await exportFailureFixture(attack);
await expect(f.prepared.restoreWorkspace()).rejects.toThrow();
expect(f.counts.fullExports).toBe(0);
expect(f.metrics).toEqual([]);
await expect(fs.stat(path.join(f.root, "new"))).rejects.toMatchObject({ code: "ENOENT" });
expect(await fs.readFile(path.join(f.root, "deleted"), "utf8")).toBe("delete me");
expect(await fs.readFile(f.outside, "utf8")).toBe("private");
await f.prepared.cleanupWorkspaceSnapshot();
}, 30000);
it.each(["full_root", "full_special"] as const)("rejects unsafe %s data before host writes after capacity fallback", async attack => {
const f = await exportFailureFixture(attack);
await expect(f.prepared.restoreWorkspace()).rejects.toThrow("Unsafe workspace export directory");
expect(f.counts).toEqual({ captures: 1, fullExports: 1, capacityFailures: 1 });
await expect(fs.stat(path.join(f.root, "new"))).rejects.toMatchObject({ code: "ENOENT" });
expect(await fs.readFile(path.join(f.root, "deleted"), "utf8")).toBe("delete me");
expect(await fs.readFile(f.outside, "utf8")).toBe("private");
await f.prepared.cleanupWorkspaceSnapshot();
}, 30000);
// RC-50b: models the Daytona provider. Every transfer stages its own
// `.paperclip-upload-*` scratch at the workspace root, and a SIGKILL during a
// transfer leaves it there. An ordinary export drops `._*` and
// `.paperclip-upload*` names at any depth and extracts under the host umask;
// a preserveSourceEntries export copies the source exactly.
it("restores a sparse checkpoint after a killed transfer left provider scratch, keeping exact entries", async () => {
const { prepareSandboxManagedRuntime } = await import("./sandbox-managed-runtime.js");
const f = await fixture(); await disposeDirectorySnapshot(f.baseline);
const remote = path.join(f.temp, "remote");
const reserved = /^(\._|\.paperclip-upload)/;
const run = async (command: string) => { await execute("sh", ["-c", command]); };
type Operations = import("./sandbox-managed-runtime.js").SandboxSyncOperation[];
const stripUmask = async (directory: string): Promise<void> => {
for (const child of await fs.readdir(directory, { withFileTypes: true })) {
const full = path.join(directory, child.name);
if (child.isDirectory()) await stripUmask(full);
else if (child.isFile()) await fs.chmod(full, (await fs.stat(full)).mode & 0o7777 & ~0o022);
}
};
const transfer = (outbound: boolean) => async (operations: Operations) => {
for (const operation of operations) {
for (const file of operation.files) {
const ordinaryExport = outbound && file.preserveSourceEntries !== true;
await fs.mkdir(path.dirname(file.targetPath), { recursive: true });
await fs.cp(file.sourcePath, file.targetPath, { recursive: true, verbatimSymlinks: true,
filter: (source) => !file.exclude?.some((exclude) => path.relative(file.sourcePath, source) === exclude)
&& !(ordinaryExport && reserved.test(path.basename(source))) });
if (ordinaryExport && (await fs.lstat(file.targetPath)).isDirectory()) await stripUmask(file.targetPath);
}
for (const command of operation.postUploadCommands ?? []) await run(command.command);
}
return { operations: operations.map((op) => ({ operationId: op.operationId, filesTransferred: 0, bytesTransferred: 0 })) };
};
const client = { makeDir: (dir: string) => fs.mkdir(dir, { recursive: true }).then(() => {}),
writeFile: (file: string, data: ArrayBuffer) => fs.writeFile(file, Buffer.from(data)), readFile: (file: string) => fs.readFile(file),
listFiles: (dir: string) => fs.readdir(dir), remove: (file: string) => fs.rm(file, { force: true, recursive: true }), run,
syncIn: transfer(false), syncOut: transfer(true) };
const prepared = await prepareSandboxManagedRuntime({ spec: { provider: "test", sandboxId: "test", remoteCwd: remote, timeoutMs: 30000, apiKey: null, transport: "sandbox" },
adapterKey: "test", client, workspaceLocalDir: f.root, workspaceCheckpoint: true });
// The s14 sandbox kept this exact file; its mtime equals the SIGKILL.
await fs.writeFile(path.join(remote, ".paperclip-upload-d99fe36b-5acc-437b-96ad-85d5ff9e79c4.tar"), Buffer.alloc(2048, 1));
await fs.writeFile(path.join(remote, "new"), "agent work");
await fs.writeFile(path.join(remote, "._summary"), "user summary");
await fs.writeFile(path.join(remote, "build.sh"), "#!/bin/sh\n"); await fs.chmod(path.join(remote, "build.sh"), 0o775);
await prepared.restoreWorkspace();
expect(await fs.readFile(path.join(f.root, "new"), "utf8")).toBe("agent work");
expect(await fs.readFile(path.join(f.root, "._summary"), "utf8")).toBe("user summary");
expect((await fs.stat(path.join(f.root, "build.sh"))).mode & 0o777).toBe(0o775);
expect((await fs.readdir(f.root)).filter(name => name.startsWith(".paperclip-upload"))).toEqual([]);
}, 30000);