fix: harden preprocessing state and child capabilities

backend/test/workspace-preprocessing-state.test.ts

@@ -1,2 +1,35 @@
-import {describe,expect,it} from "vitest"; import {mkdtemp} from "node:fs/promises"; import {join} from "node:path"; import {PreprocessingStateStore} from "../src/workspaces/preprocessing-state.js";
-describe("durable preprocessing state",()=>{it("creates and replays exact state",async()=>{const root=await mkdtemp(join(process.env.TMPDIR??"/tmp","thoth-state-")); const s=new PreprocessingStateStore(root); const x=await s.create({workspaceId:"abc-workspace" as never,revision:"a".repeat(40),operation:"schema"}); expect((await s.loadForResume({stateRoot:root,runId:x.runId,workspaceId:x.workspaceId,revision:x.revision,operation:x.operation})).runId).toBe(x.runId);});});
+import { describe, expect, it, afterEach } from "vitest";
+import { mkdtemp, readFile, chmod, writeFile, symlink, lstat } from "node:fs/promises";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import { PreprocessingStateStore } from "../src/workspaces/preprocessing-state.js";
+
+const roots: string[] = [];
+afterEach(async () => { for (const root of roots.splice(0)) await import("node:fs/promises").then(fs => fs.rm(root, { recursive: true, force: true })); });
+async function makeStore() { const root = await mkdtemp(join(tmpdir(), "thoth-state-")); roots.push(root); return { root, store: new PreprocessingStateStore(root) }; }
+const input = { workspaceId: "abc-workspace" as never, revision: "a".repeat(40), operation: "schema" };
+
+describe("durable preprocessing state", () => {
+  it("creates and replays exact state with immutable identity", async () => {
+    const { root, store } = await makeStore(); const state = await store.create(input);
+    const replay = await store.loadForResume({ ...input, runId: state.runId });
+    expect(replay).toEqual(state); await expect(store.transition(state.runId, { phase: "introspected", workspaceId: "evil" } as never)).rejects.toThrow("preprocessing_conflict");
+  });
+  it("rejects tampered, traversal, mode and version state before returning it", async () => {
+    const { root, store } = await makeStore(); const state = await store.create(input); const path = join(root, "preprocessing", "jobs", `${state.runId}.json`);
+    const original = JSON.parse(await readFile(path, "utf8")); await writeFile(path, JSON.stringify({ ...original, schemaVersion: 9 }));
+    await expect(store.load({ ...input, runId: state.runId })).rejects.toThrow();
+    await writeFile(path, JSON.stringify(original)); await chmod(path, 0o644); await expect(store.load({ ...input, runId: state.runId })).rejects.toThrow("preprocessing_conflict");
+    await expect(store.load({ ...input, runId: "../" + state.runId })).rejects.toThrow();
+  });
+  it("writes candidate artifacts atomically with bounded bytes and immutable links", async () => {
+    const { root, store } = await makeStore(); const state = await store.create(input); const bytes = new TextEncoder().encode("tables: []\n");
+    const artifact = await store.writeFkCandidate(state.runId, bytes); expect(artifact.bytes).toBe(bytes.byteLength);
+    const candidate = lstat(join(root, "preprocessing", "fk-candidates", `${state.runId}.yaml`)); expect((await candidate).nlink).toBe(1);
+    await expect(store.writeFkCandidate(state.runId, new Uint8Array(1 << 20))).rejects.toThrow("preprocessing_conflict");
+  });
+  it("rejects a symlinked state root", async () => {
+    const real = await mkdtemp(join(tmpdir(), "thoth-state-real-")); const link = join(tmpdir(), `thoth-state-link-${Date.now()}`); roots.push(real, link); await symlink(real, link);
+    expect(() => new PreprocessingStateStore(link)).toThrow("preprocessing_conflict");
+  });
+});