Skip to content

Microsoft-only packages, AI attribution, and a guide for AI reviewers - #4

Merged
edwest19 merged 2 commits into
mainfrom
chore/pure-dotnet-ai-attribution
Nov 5, 2026
Merged

edwest19 merged 2 commits into
mainfrom
chore/pure-dotnet-ai-attribution

Conversation

@edwest19

@edwest19 edwest19 commented Nov 5, 2026

Copy link
Copy Markdown
Owner

What this does

Housekeeping for the project's ground rules. No app behavior changes: the edits to app code are comments only, and every password stays the same.

  1. No third-party packages. The tests now use MSTest, Microsoft's own test framework, instead of xUnit. The only packages left in the whole repo are Microsoft's:

    Package Used by Why
    Microsoft.WindowsAppSDK SecretKeyUI WinUI 3 window
    Microsoft.Windows.SDK.BuildTools SecretKeyUI MSIX packaging
    MSTest tests only test framework
  2. Rules that check themselves. A new RepositoryRulesTests fails if:

    • a package that isn't on that list is added
    • app code uses any network API (HttpClient, sockets, System.Net, WebView…)
    • the MSIX manifest asks for any permission other than runFullTrust
  3. Removed an unused permission. The manifest asked for systemAIModels (Windows' on-device AI models), which the app never uses. It's gone.

  4. AI-generated code is marked. Every file Claude wrote or changed has a header comment saying so. In existing files, the exact lines Claude added are marked too. Commits keep their Co-Authored-By: Claude line.

  5. README "About this project" states the purpose (an app built with AI-generated code), that it's local only, and the package policy.

  6. CLAUDE.md is a guide that any AI assistant reads first. It covers:

    • the ground rules
    • the folder layout
    • how to build and test
    • a step-by-step way to check every README statement against the code
    • the README statements that are known to be wrong today

How it was checked

  • All 37 tests pass with a local stand-in runner (NuGet is blocked where this was written); the test check below is the real MSTest run.
  • Each new rule test was proven to fail when its rule is broken: adding a third-party package, adding an HttpClient line, and adding an internet permission each made exactly one test fail.
  • The MSTest assertions used were confirmed against Microsoft's MSTest 4.0.1 source.

Next

Pull request #5 fixes the README gaps listed in CLAUDE.md: passwordmask=, column handling, multi-line notes, settings validation. It also adds tests that tie each README claim to code.


Generated by Claude Code

- Tests: switch from xUnit (third-party) to MSTest 4.0.1 (Microsoft).
  Same 34 tests and expected values.
- New RepositoryRulesTests enforce the ground rules:
  - only allowed Microsoft packages
  - no network APIs in app code
  - manifest declares only runFullTrust
- Manifest: remove the unused systemAIModels capability.
- Mark AI-generated code: header comments on every file Claude wrote or
  changed, and inline markers at Claude's changes in existing files.
- README: "About this project" states the purpose (an app built with
  AI-generated code), local-only operation and the package policy.
- CLAUDE.md: rules, layout, build/test, how an AI should check the
  README against the code, and the known README gaps to fix next.

No app behavior changes; app code edits are comments only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QSQZn3X5gDJdRjbqMfKRpd
Addresses the two MSTEST0037 analyzer warnings from CI
(IsNotEmpty and IsGreaterThan instead of IsTrue).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QSQZn3X5gDJdRjbqMfKRpd
@edwest19
edwest19 merged commit 6fc0ece into main Nov 5, 2026
2 checks passed
@edwest19
edwest19 deleted the chore/pure-dotnet-ai-attribution branch November 6, 2026 03:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants