Add end-to-end coverage for LibreSign AI integrations

Đang mở
#9,060 0 bình luận 0 reaction 0 người được giao Xem trên GitHub

Maintainer thường phản hồi trong vòng 1 ngày

Chưa có ai nhận issue này.

Đánh giá

Độ khó
5/5
Thời gian dự kiến
Hơn một tuần
Mức phù hợp với người mới
35/100
Loại issue
Tính năng
Độ rõ ràng
Khá rõ ràng
Mức độ hoạt động
Sôi nổi
Công nghệ
php
Lĩnh vực
ai, testing

Hướng nghiên cứu

Start by reading the AI foundation epic and the existing LibreSign Task Processing integration and tests; the issue names no specific files or test entry points. Map the shared AI path and existing Nextcloud AI test patterns, then add deterministic coverage first and assess whether provider-contract and local-model smoke tests add unique value. Done means normal PR CI covers stable shared-path and security contracts without paid APIs or assertions on generated prose.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.

Mô tả

Goal

Add layered end-to-end and integration coverage that proves LibreSign AI features work through the real Nextcloud integration path without making ordinary CI depend on paid APIs or unstable generated prose.

This issue is part of the AI foundation epic.

Test layers

Use three complementary layers.

1. Deterministic pull-request coverage

Use the LibreSign deterministic Task Processing provider/test double for fast, exact and reproducible behavior.

Cover representative end-to-end flows such as:

  • capability unavailable;
  • capability available and policy allowed;
  • capability available and policy denied;
  • successful task/result;
  • failure and retry/recovery behavior;
  • invalid/hostile output;
  • relevant cancellation/asynchronous behavior;
  • frontend visibility/action state follows backend capability and policy.

This layer should be suitable for normal pull-request CI.

2. Provider integration contract

Where feasible, exercise LibreSign through an official Nextcloud provider integration against a deterministic local compatible endpoint/test backend.

The purpose is to cover:

LibreSign -> OCP Task Processing -> provider integration -> provider protocol -> Task Processing result

without paid/network-dependent model calls.

Only add this layer if it materially covers behavior that the deterministic provider cannot.

3. Real local-AI smoke coverage

Use a small number of real local-provider smoke tests where this adds confidence.

Follow the patterns used by Nextcloud AI applications such as llm2:

https://github.com/nextcloud/llm2

Use CPU execution/model caching where practical and avoid making generated wording an assertion.

Real-model tests should check stable invariants such as:

  • task completes successfully;
  • expected output shape is present;
  • file/document input reaches the supported task path;
  • structured output can be parsed/validated;
  • security/policy invariants remain enforced.

Initially this layer may run on relevant paths, main/nightly or a dedicated workflow if its runtime is not appropriate for every pull request.

Context Chat precedent

For future retrieval work, follow the ACL/lifecycle style of the official Context Chat integration tests:

https://github.com/nextcloud/context_chat

That includes proving content is unavailable without access and removed from retrieval after access/delete changes.

Do not implement future Context Chat tests in this foundation issue before that feature exists.

Flake/cost policy

  • No paid API requirement for required CI.
  • No assertion on exact natural-language output from a real model.
  • Pin/cache model/test infrastructure where appropriate.
  • Keep real-model smoke scenarios few and contract-focused.
  • A provider outage must not make normal pull-request CI nondeterministic.

Security scenarios

Ensure the layered test architecture can exercise:

  • prompt-injection-like document content;
  • malformed structured output;
  • unsafe generated markup where a feature consumes markup;
  • cross-user task/result access;
  • policy denial;
  • provider absence/failure.

Feature-specific adversarial cases remain the responsibility of the feature's implementation issues.

Done when

  • Normal PR CI has deterministic end-to-end coverage of the shared AI path.
  • The Task Processing boundary is exercised beyond unit mocks.
  • Provider integration is covered where it adds unique value.
  • A focused real local-AI smoke path exists where feasible and justified.
  • Required CI does not depend on paid external AI.
  • Tests assert stable contracts rather than exact real-model prose.
  • Security/authorization failure paths are represented in the test architecture.
Ngôn ngữ chính
PHP
Star
844
Fork
176
Merge trung bình
14 giờ 3 phút
Pull request đã merge (30 ngày)
344

Chuẩn bị môi trường

Mở trong Codespaces

Khởi chạy dev container của dự án ngay trên trình duyệt, bằng tài khoản GitHub của bạn.

Bắt đầu từ đâu

  1. Đọc hết issue, rồi đọc hướng dẫn đóng góp của dự án.
  2. Bình luận trên issue rằng bạn sẽ nhận — tránh hai người làm cùng một việc.
  3. Fork repository và làm thay đổi trên một nhánh.
  4. Mở pull request có tham chiếu số hiệu của issue.

Issue khác của LibreSign/libresign

Tất cả issue của LibreSign/libresign

Issue tương tự

Thêm issue về PHP

Nhận issue mới trong hộp thư của bạn

Bản tóm tắt ngắn những issue GitHub phù hợp với người mới.