GHSA-pp95-gc86-jq6q

### Summary

The run replay `action` function at `apps/webapp/app/routes/resources.taskruns.$runParam.replay.ts` has no authentication or authorization check. While the `loader` (GET) in the same file properly calls `requireUser(request)` and scopes queries to the user's organizations, the `action` (POST) at line 166 does neither — allowing any authenticated user to replay task runs from any organization by knowing the run's `friendlyId`.

### Details

**Vulnerable file:** `apps/webapp/app/routes/resources.taskruns.$runParam.replay.ts`

**The `loader` (line 28-29) — properly authenticated:** ```typescript export async function loader({ request, params }: LoaderFunctionArgs) { const user = await requireUser(request); // ✓ Auth check const userId = user.id; // ... queries scoped to user's orgs } ```

**The `action` (line 166-193) — NO authentication:** ```typescript export const action: ActionFunction = async ({ request, params }) => { const { runParam } = ParamSchema.parse(params); // ✗ NO requireUser() call // ✗ NO requireUserId() call // ✗ NO org membership check

const taskRun = await prisma.taskRun.findFirst({ where: { friendlyId: runParam, // Queries ANY run, no org scoping }, include: { runtimeEnvironment: { select: { slug: true } }, project: { include: { organization: true } }, }, });

// ... proceeds to replay the run in the victim's environment const replayRunService = new ReplayTaskRunService(); ```

The Prisma query at line 177 fetches the run by `friendlyId` only — no `userId` or organization filter. The `ReplayTaskRunService` then creates a new task run in the victim's environment, executing with the victim's environment variables and secrets.

**Same bug class exists in:** `apps/webapp/app/routes/resources.batches.$batchId.check-completion.ts` (line 17) — the `action` has zero authentication, allowing any user to trigger batch completion for any batch ID.

### PoC

**Run replay IDOR:** ```bash # Any authenticated user can replay any org's task run POST /resources/taskruns/run_abc123def/replay Cookie: <any-valid-session> Content-Type: application/x-www-form-urlencoded

environmentId=<victim-env-id>&failedRedirect=/ ```

The `friendlyId` values (e.g., `run_abc123def`) are short, incrementing strings that can be enumerated.

**Batch completion (same bug class):** ```bash # Any authenticated user can trigger batch completion for any batch POST /resources/batches/<batchId>/check-completion Cookie: <any-valid-session> Content-Type: application/x-www-form-urlencoded

redirectUrl=/ ```

### Impact

- **Cross-organization task execution:** An attacker can replay task runs belonging to other organizations, executing tasks in the victim's environment with the victim's secrets and API keys - **Secret exposure:** Replayed tasks run with the victim organization's environment variables, which may contain database credentials, API keys, and other secrets - **Resource consumption:** Attacker consumes the victim's compute quota by replaying their tasks - **Data integrity:** The batch completion endpoint can prematurely resume parent tasks waiting for batch results, causing data integrity issues - **Low attack complexity:** `friendlyId` values are short, predictable strings — enumeration is feasible