Add AI memory-leak scanner + fixer agentic workflows

[kubaflo]

Note

Are you waiting for the changes in this PR to be merged?
It would be very helpful if you could test the resulting artifacts from this PR and let us know in a comment if this change resolves your issue. Thank you!

Note

🤖 AI-generated PR. These workflows, and this description, were produced with AI assistance. Please review carefully before merging.

Summary

Adds two gh-aw agentic workflows that proactively find and fix managed, cross-platform memory leaks in MAUI. They mirror the AI-driven leak-hunting flow several contributors already run by hand (scan → repro → measure → file → fix) and turn it into a repeatable, reviewable pipeline.

Key property: neither the detection pass nor the fix-validation pass needs an emulator/simulator. The leak signature is proven by a plain dotnet test on the .NET library TFM, so both run on a standard GitHub-hosted runner.

Only 4 files are added — the workflow source (.md) and its compiled artifact (.lock.yml) for each:

.github/workflows/daily-leak-hunter.md   + .lock.yml
.github/workflows/leak-fixer.md          + .lock.yml

---

1. daily-leak-hunter — the scanner (files issues)

Trigger: schedule: every 12h + manual (workflow_dispatch). Files at most one issue per run.

Pass A — runtime leak hunt → [leak-scan] issue

1. Picks a rotating focus area of the managed surface (src/Core/src, src/Controls/src/Core, src/Essentials/src).
2. Looks for the classic leak signature: a long-lived / shared / static root holding a strong reference to a transient (control / view-model / handler) with no teardown on unload.
3. Writes a standalone control / leaky / mitigation xUnit repro that references the shipped Microsoft.Maui.Controls NuGet package (no source build), and measures retention with WeakReference + a forced GC.
4. Runs it. Files a [leak-scan] issue only if the leak is empirically confirmed — the leaky scenario retains while both the control and the mitigation release. A false positive is treated as worse than a quiet run, so unconfirmed candidates are dropped.

Pass B — leak-test coverage-gap scan → [leak-test-gap] issue

Runs only when Pass A files nothing (the expected case once the runtime surface is hardened). It statically diffs MAUI's concrete control surface against its existing memory-leak test coverage (src/Controls/tests/DeviceTests/Memory/MemoryTests.cs HandlerDoesNotLeak [InlineData] list + per-control *DoesNotLeak tests + the Appium MemoryTests) and files a [leak-test-gap] issue proposing the precise missing test, with a copy-paste-ready stub. This keeps every run productive and closes the holes through which future leaks slip in untested.

De-dup & safety

• Skips a candidate already covered by one of its own open [leak-scan] / [leak-test-gap] issues.
• Never touches product or test code (read-only + one create-issue).
• noop: report-as-issue: false — quiet runs don't create noise.

---

2. leak-fixer — the fixer (opens draft PRs)

Trigger: schedule: every 12h + manual (workflow_dispatch). At most one action per run, with de-dup + a 3-attempt cap. Each run it picks the highest-priority of three tracks:

Track C — respond to review feedback (HIGHEST priority, checked first)

Before doing new work, it checks whether one of its own open [leak-fix]/[leak-test] PRs has an unaddressed CHANGES_REQUESTED review. If so, it works on that PR: applies the valid requested changes (pushes a commit to the PR branch, re-validated so the PR's own test still holds) and/or posts a comment pushing back on any request that is wrong, would regress behaviour, or asks for a mute. A loop-guard (only act on a review newer than the PR's last commit and the workflow's last comment) means it never re-processes the same review. Uses the push-to-pull-request-branch + add-comment safe-outputs, scoped by required-title-prefix: "[leak-" to its own PRs. Example this is built for: MauiBot requested changes on #36253 (a real ❌ duplicate-MergedDictionaries unsubscribe bug + suggestions) — Track C applies the sound fixes and replies re: any it disagrees with.

Track A — [leak-scan] (runtime leak → [leak-fix] PR)

For the selected leak it:

1. Writes a focused regression test in Controls.Core.UnitTests.
2. Builds MAUI from source and confirms the test FAILS on main — proving it catches the leak.
3. Implements the minimal, idiomatic managed fix — a weak subscription / teardown mirroring the existing WeakEventManager / WeakNotify*Proxy / BindableLayoutController patterns.
4. Rebuilds and confirms the same test now PASSES, no neighbouring regressions.
5. Opens a draft [leak-fix] PR with Fixes #N and the red→green evidence.

Track B — [leak-test-gap] (missing coverage → [leak-test] PR)

For a coverage gap it adds the memory-leak test the issue proposes (test-only, no product change): a runner-run Controls.Core.UnitTests guard and/or a HandlerDoesNotLeak [InlineData] in MemoryTests.cs. It validates as far as the runner allows — a Core.UnitTests guard must build and PASS on main (device-test [InlineData]s can't run on the runner, so they're validated structurally and execute on maui-pr-devicetests) — then opens a draft [leak-test] PR. If a runnable guard unexpectedly fails, that's a real live leak → the fixer escalates to Track A and writes the fix instead.

Safety

• Enforces red→green in both directions (Track A) — a fix without a demonstrated failing-then-passing test is rejected.
• Never mutes / skips / disables a test; rejects its own attempt if the only thing that goes green is a mute — including when a review asks for one (it pushes back with a comment instead).
• One action per run — Track C review-response or one new Track A/B PR, never both.
• Writes only via scoped safe-outputs (agent job is read-only); Track C can only touch PRs whose title starts with [leak-.
• Skips cleanly if already fixed on main, out of scope (needs a device fix), or attempt-capped.

---

Examples already produced

These are real, human-reviewable outputs the workflows generated automatically.

End-to-end, running on dotnet/maui itself (pre-merge validation)

Both workflows were executed on this repo's own Actions infrastructure (via short-lived
push-triggered test branches, since workflow_dispatch only lights up once a workflow is on
main). Every run succeeded; the outputs below are all on dotnet/maui:

Run	Mode	Output	What it proves
hunter	Pass B	issue #36306 — [leak-test-gap] FlexLayout has no HandlerDoesNotLeak test	finds a genuinely-uncovered control
fixer	Track B	PR #36307 — [leak-test] Add memory-leak coverage for FlexLayout (test-only: a Core.UnitTests guard and the device [InlineData])	closes a coverage gap
hunter	Pass A	issue #36308 — [leak-scan] Rooted ResourceDictionary merged into a transient element	empirically-confirmed runtime leak
fixer	Track A	PR #36309 — [leak-fix] Fix ResourceDictionary MergedDictionaries memory leak (real product fix in WeakEventProxy.cs + ResourceDictionary.cs, source-built + red→green)	produces a validated fix from source
hunter	Pass B	issue #36310 — [leak-test-gap] No test asserts Shell itself is released	finds another uncovered control

This exercised all four paths on the real repo: Pass A (find leak), Pass B (find test
gap), Track A (fix leak), Track B (add coverage) — with zero run failures.

De-dup note (tuning item): #36308/#36309 re-covers the same leak as #36251/#36253 below,
because on this repo the hunter currently de-dups only against its own [leak-scan]-titled
issues and doesn't yet recognize the promoted issues / existing [leak-fix] PRs. Broadening
the de-dup to any open perf/memory-leak issue + open leak PR removes this — happy to add it.

Fix PRs the leak-fixer opened earlier (fork-built, promoted here)

Each carries a regression test that fails on main without the fix and passes with it, plus the idiomatic weak-proxy fix. All are the same real leak family — a shared/long-lived value retaining a control via a non-weak subscription with no unload teardown:

Leak (rooting API)	Issue	Fix PR
IndicatorView.ItemsSource	#35775	#36252
ResourceDictionary.ValuesChanged	#36251	#36253
Border.StrokeDashArray	#35492	#36254
Path.Data / Path.RenderTransform	#35860	#36273
Picker.ItemsSource	#36272	#36274

(#36251 and #36272 are the issues this pipeline filed; #35775 / #35492 / #35860 were pre-existing reports the fixer linked to instead of duplicating.)

Raw scanner output (on the validation fork)

• [leak-scan] (Pass A) — e.g. [leak-scan] Repeating animations leak the animated VisualElement via the static AnimationExtensions.s_animations dictionary kubaflo/maui#86 (repeating animations), [leak-scan] SwipeItemView.Command leaks the bound view via non-weak ICommand.CanExecuteChanged subscription with no teardown kubaflo/maui#88 (SwipeItemView.Command), [leak-scan] ListView.RefreshCommand leaks the ListView via non-weak ICommand.CanExecuteChanged subscription with no teardown kubaflo/maui#89 (ListView.RefreshCommand).
• [leak-test-gap] (Pass B) — e.g. [leak-test-gap] FlexLayout has no memory-leak (HandlerDoesNotLeak) test kubaflo/maui#95 (FlexLayout has no HandlerDoesNotLeak test), [leak-test-gap] AbsoluteLayout has no memory-leak (HandlerDoesNotLeak) test kubaflo/maui#96 (AbsoluteLayout), [leak-test-gap] SwipeItemView has no memory-leak (DoesNotLeak) device test kubaflo/maui#103 (SwipeItemView).

Across the fork run the hunter filed ~18 confirmed [leak-scan] + ~10 [leak-test-gap] issues, and the fixer converted several into the validated PRs above.

---

Setup required — essentially none

These workflows add no new secrets, environments, or labels. They reuse the exact infrastructure the repo's existing agentic workflows already run on (ci-status-main, ci-status-fix, daily-repo-status, agentic-labeler, copilot-evaluate-tests, rerun-review-scanner):

Requirement	Status	Detail
Copilot engine token	✅ already configured	Both use environment: gh-aw-agents — the same environment all six current agentic workflows use. The referenced secret set is byte-identical to ci-status-fix (COPILOT_GITHUB_TOKEN, GH_AW_*, GITHUB_TOKEN). No new secret.
agentic-workflows label	✅ already exists	Used by the create-issue / create-pull-request safe-outputs.
Actions may open PRs (fixer)	✅ already enabled	ci-status-fix already opens PRs on this repo, so "Allow GitHub Actions to create and approve pull requests" is already on. Nothing to toggle.
Network egress	✅ declared in-workflow	The dotnet ecosystem in each file allowlists nuget.org (hunter) and the dnceng feeds pkgs.dev.azure.com (fixer's from-source restore). No repo/org change.
Job permissions	✅ declared in-workflow	Agent jobs are read-only; writes happen only in gh-aw's scoped safe-output jobs.

So on this repo they're ready to run as soon as they land on the default branch — no environment or config changes needed. (The only place any setup would be required is a fork without the gh-aw-agents environment, where you'd add a single COPILOT_GITHUB_TOKEN repo secret.)

Triggering / editing

• Run from the Actions tab, or gh workflow run "daily-leak-hunter.lock.yml" / "leak-fixer.lock.yml". The fixer accepts issue_number (scope to one issue) and dry_run (do everything, emit no PR) inputs.
• The .lock.yml is what Actions runs; the .md is the source. After any .md edit run gh aw compile <name> and commit the regenerated .lock.yml. Compiled with the repo's pinned gh-aw v0.79.8, so actions-lock.json is unchanged.

Conventions

Adapted to match the existing ci-status-* agentic workflows: environment: gh-aw-agents, if: github.repository == 'dotnet/maui', agentic-workflows label, dotnet network ecosystem, safe-output-scoped writes.

Cadence (please tune)

The hunter is scheduled every 12h (matching ci-status-main). Both are scheduled every 12h (matching ci-status-main / ci-status-fix) and also runnable manually. The fixer works through the hunter's backlog (leaks first, then coverage gaps), one draft PR per run, with de-dup + a 3-attempt cap. Happy to change either interval — whatever the team prefers.

[github-actions]

🚀 Dogfood this PR with:

⚠️ WARNING: Do not do this without first carefully reviewing the code of this PR to satisfy yourself it is safe.

curl -fsSL https://raw.githubusercontent.com/dotnet/maui/main/eng/scripts/get-maui-pr.sh | bash -s -- 36276

Or

• Run remotely in PowerShell:

iex "& { $(irm https://raw.githubusercontent.com/dotnet/maui/main/eng/scripts/get-maui-pr.ps1) } 36276"

[bcaceiro]

Cool concept! Would be also interesting to create one for performance!

[jonathanpeppers]

Could we revive this PR?

• #18318

It wouldn't need AI, but you could use AI to fix all the warnings that the deterministic analyzer finds.

[kubaflo]

Note

🤖 AI-generated comment (GitHub Copilot CLI, on behalf of @kubaflo).

Superseded by #36342, which hosts the same work on a dotnet/maui branch (feature/memory-leak-scanner-workflows) instead of a fork — full-trust CI and the agentic workflows can run natively. Continuing there. Closing this one.