Wysoka

praisonai-platform: Label endpoints' unchecked label_id/issue_id enable cross-workspace label IDOR (edit, delete, link)

  • Ujawniono: 29.05.2026
  • Aktualizacja: 11.10.2026 14:18

Summary

Type: Insecure Direct Object Reference. Five label endpoints — PATCH /workspaces/{workspace_id}/labels/{label_id}, DELETE .../labels/{label_id}, POST .../issues/{issue_id}/labels/{label_id}, DELETE .../issues/{issue_id}/labels/{label_id}, GET .../issues/{issue_id}/labels — gate access on require_workspace_member(workspace_id) only and pass URL-supplied label_id and issue_id straight through to LabelService without verifying either belongs to the workspace. File: src/praisonai-platform/praisonai_platform/services/label_service.py, lines 35-100; route handlers at src/praisonai-platform/praisonai_platform/api/routes/labels.py, lines 42-106. Root cause: identical pattern to the agent / issue / project / comment IDORs in this codebase: the route's workspace_id is used as a membership predicate but never threaded through to the service layer. LabelService.get(label_id) runs session.get(IssueLabel, label_id) with no workspace filter; update/delete inherit the gap; add_to_issue(issue_id, label_id) and remove_from_issue(issue_id, label_id) write/delete association rows without verifying either ID belongs to the membership-checked workspace; list_for_issue(issue_id) reads them.

Affected Code

File 1: src/praisonai-platform/praisonai_platform/services/label_service.py, lines 35-100.

python
class LabelService:
    ...

    async def get(self, label_id: str) -> Optional[IssueLabel]:
        return await self._session.get(IssueLabel, label_id)         # <-- BUG: no workspace_id predicate

    async def update(
        self,
        label_id: str,
        ...
    ) -> Optional[IssueLabel]:
        label = await self.get(label_id)                             # <-- inherits the gap
        ...

    async def delete(self, label_id: str) -> bool:
        label = await self.get(label_id)                             # <-- inherits the gap
        ...

    async def add_to_issue(self, issue_id: str, label_id: str) -> None:
        # writes a row in issue_label association table; no workspace check on either id

    async def remove_from_issue(self, issue_id: str, label_id: str) -> None:
        # deletes from association table; no workspace check on either id

    async def list_for_issue(self, issue_id: str) -> list[IssueLabel]:
        # reads from association table; no workspace check on issue_id

File 2: src/praisonai-platform/praisonai_platform/api/routes/labels.py, lines 42-106.

python
@router.patch("/labels/{label_id}", response_model=LabelResponse)
async def update_label(workspace_id: str, label_id: str, body: LabelUpdate, ...):
    svc = LabelService(session)
    label = await svc.update(label_id, body.name, body.color)        # <-- writes any label in the DB
    ...

@router.delete("/labels/{label_id}", ...)
async def delete_label(workspace_id: str, label_id: str, ...):
    deleted = await svc.delete(label_id)                             # <-- deletes any label in the DB
    ...

@router.post("/issues/{issue_id}/labels/{label_id}", ...)
async def add_label_to_issue(workspace_id: str, issue_id: str, label_id: str, ...):
    await svc.add_to_issue(issue_id, label_id)                       # <-- attaches any label to any issue cross-workspace

@router.delete("/issues/{issue_id}/labels/{label_id}", ...)
async def remove_label_from_issue(workspace_id: str, issue_id: str, label_id: str, ...):
    await svc.remove_from_issue(issue_id, label_id)                  # <-- detaches any label from any issue cross-workspace

@router.get("/issues/{issue_id}/labels", ...)
async def list_issue_labels(workspace_id: str, issue_id: str, ...):
    labels = await svc.list_for_issue(issue_id)                      # <-- reads label assignments for any issue

Why it's wrong: the workspace_id URL segment is treated as a UI hint; the actual label_id and issue_id lookups query the database without a workspace constraint. The MemberService in this same codebase uses a composite key correctly; the label service does not. The add_to_issue and remove_from_issue paths are particularly nasty because they touch two unverified IDs at once: an attacker can attach a foreign workspace's label to a foreign workspace's issue (or detach the legitimate labels), corrupting both sides of an association the attacker has no business touching.

Exploit Chain

  1. Attacker registers a workspace W_attacker (member) and harvests a foreign-workspace label_id L_T and a foreign-workspace issue_id I_T. Both leak via list_labels responses (which include label IDs — but only for W_attacker; for the target the IDs come from issue records that include label associations, activity feeds, exported dumps, error messages). State: attacker holds L_T and I_T.
  2. Attacker authenticates and sends PATCH /workspaces/W_attacker/labels/L_T with {"name": "<deleted>", "color": "#000000"}. require_workspace_member(W_attacker, attacker) passes. LabelService.update(L_T, ...) loads the foreign label and renames it. State: every issue across the foreign workspace that bears this label now displays the attacker-chosen name and colour.
  3. Attacker sends DELETE /workspaces/W_attacker/labels/L_T. LabelService.delete(L_T) deletes the foreign label, dropping every issue-label association row that referenced it (cascade or orphan, depending on schema). State: foreign workspace's labels are gone or corrupted.
  4. Attacker sends POST /workspaces/W_attacker/issues/I_T/labels/L_T2 to attach foreign label L_T2 to foreign issue I_T. LabelService.add_to_issue(I_T, L_T2) writes the association row regardless of either ID's workspace. State: the foreign issue now carries an arbitrary attacker-chosen label, which surfaces in every filter/search/board view in the foreign workspace's UI.
  5. Attacker sends DELETE /workspaces/W_attacker/issues/I_T/labels/L_legit to strip the legitimate label off the foreign issue. State: triagers can no longer find the issue via label filters.
  6. Attacker sends GET /workspaces/W_attacker/issues/I_T/labels to read the current label set on any foreign issue. State: the attacker fingerprints the foreign workspace's triage taxonomy.
  7. Final state: with one workspace-member token plus harvested foreign IDs, the attacker rewrites and deletes other workspaces' labels, attaches/detaches arbitrary labels on other workspaces' issues, and reads triage state across the deployment.

Security Impact

Severity: sec-moderate. CVSS 6.3: network attack, low complexity, low privileges, no user interaction, scope unchanged. The integrity damage is high (rename/delete of foreign labels is permanent and silent; cross-workspace label-attachment corrupts UI filters), confidentiality is low (label names are not the most sensitive field but do leak triage taxonomy), availability low (foreign workspaces may lose triage visibility into their own issues until the labels are restored). Attacker capability: rename and delete any label in the multi-tenant deployment; attach any label to any issue; detach any label from any issue; list label assignments for any issue. Combined with the companion IssueService IDOR (separate advisory), the attacker can also modify the underlying issue, making the cross-workspace tampering very difficult to detect. Preconditions: praisonai-platform is deployed multi-tenant; the attacker has any membership token; target IDs are known or guessable. Differential: source-inspection-verified end-to-end. The asymmetry between LabelService.list_for_workspace(workspace_id) (correctly workspace-scoped) and LabelService.get(label_id) / add_to_issue(issue_id, label_id) (no workspace check) confirms the gap. With the suggested fix below, label and issue IDs that do not belong to the membership-checked workspace return 404, and the attacker cannot touch them.

Suggested Fix

Make every single-row label lookup take the workspace predicate; verify both issue_id and label_id belong to workspace_id for the association routes.

diff
--- a/src/praisonai-platform/praisonai_platform/services/label_service.py
+++ b/src/praisonai-platform/praisonai_platform/services/label_service.py
@@ -33,7 +33,12 @@ class LabelService:
         return label

-    async def get(self, label_id: str) -> Optional[IssueLabel]:
-        return await self._session.get(IssueLabel, label_id)
+    async def get(self, workspace_id: str, label_id: str) -> Optional[IssueLabel]:
+        stmt = select(IssueLabel).where(
+            IssueLabel.id == label_id,
+            IssueLabel.workspace_id == workspace_id,
+        )
+        return (await self._session.execute(stmt)).scalar_one_or_none()

-    async def add_to_issue(self, issue_id: str, label_id: str) -> None:
+    async def add_to_issue(self, workspace_id: str, issue_id: str, label_id: str) -> None:
+        # Verify both ids belong to workspace_id before writing the association row.

Then update the route handlers in routes/labels.py to thread workspace_id through every call. The same single-key-lookup pattern is filed separately for AgentService, IssueService, ProjectService, and CommentService — each is its own exploitable IDOR.

Czego dotyczy

Kliknij technologię, aby zobaczyć wszystkie dotyczące jej podatności.

PlatformaPaczkaPodatne wersjePoprawka w wersji
Python (PyPI) praisonai-platform < 0.1.4 0.1.4

Odnośniki

Ocena CVSS 7.6/10
Wektor CVSS
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:H/A:L
Słabość (CWE)
CWE-639
Platforma
Python (PyPI)
Paczka
praisonai-platform
Ujawniono
29.05.2026
Aktualizacja
11.10.2026 14:18

Źródła i licencje

OSV / GitHub Advisory Database - PYSEC-2026-2929 - PyPA Advisory Database; CC BY 4.0

OSV / GitHub Advisory Database - GHSA-5jx9-w35f-vp65 - CC BY 4.0

Historia podatności: praisonai-platform

Wszystkie opisane podatności tej paczki (Python (PyPI)), od najnowszej. Widać, czy to jednorazowa wpadka, czy paczka, której trzeba pilnować.

Wszystkie (24) w archiwum
Wszystkich wpisów
24
Pierwsza → ostatnia
05.2026 10.2026
Bez poprawki
0
Wykorzystywana w atakach
0

Rozkład istotności

  • Krytyczna 6
  • Wysoka 13
  • Średnia 5

Ujawnienia w latach

  1. 2026: 24
Istotność:
  1. praisonai-platform: Label endpoints' unchecked label_id/issue_id enable cross-workspace label IDOR (edit, delete, link) ten wpis
    Wysoka · 7.6 CVE-2026-47414 < 0.1.4 0.1.4

Dowiedz się, które podatności dotyczą Twojej aplikacji

Po podłączeniu integracji DockRay i włączeniu raportowania oprogramowania DockRay porównuje zgłoszone wersje komponentów ze znanymi podatnościami i powiadamia o potwierdzonych dopasowaniach. Zakres zależy od obsługiwanej technologii i dostępnych biuletynów.

Podłącz swoją aplikację
Newsletter DockRay

Bezpieczeństwo aplikacji w Twojej skrzynce

Najważniejsze podatności, konkretne zalecenia i praktyczne wskazówki dla Twoich aplikacji. Wybierz technologie i śledź to, co ma znaczenie dla Twojego projektu.

  • Codzienne zestawienie nowych podatności, także krytycznych.
  • Powiadomienia o ważnych aktualizacjach.
  • Praktyczne porady o bezpieczeństwie i monitoringu aplikacji.

Pozostaw puste, aby otrzymywać informacje o wszystkich technologiach.

Zapis jest bezpłatny. Wypiszesz się jednym kliknięciem w wiadomości.