Ensure the plugin's custom UI components are accessible to users with screen readers and other assistive technologies.

Add a new unit test for a class that currently lacks one or add a new test case to an existing test file, verifying improvement with Kover.

Use when writing, editing, refactoring, or debugging tests anywhere under `tests/**` (unit, integration, or end-to-end); authoring NUnit 4 test classes (`*Test`, `*IntegrationTest`); using NSubstitute mocks, AutoFixture with `[AutoMockData]` / `[InlineAutoMockData]` / `[Frozen]`, or AwesomeAssertions (NOT FluentAssertions); working with `CliIntegrationFixture`, `IntegrationTestFixture`, `MockFileSystem`, `TestableLogger`, `NUnitAnsiConsole`, or `New*` factory helpers; adding or updating E2E fixtures under `tests/Recyclarr.EndToEndTests/Fixtures/` (`recyclarr.yml`, `settings.yml`, `custom-formats-*`, `trash-guides-override/`); running `Run-E2ETests.ps1` or `scripts/coverage.py`; investigating flaky tests, coverage gaps, or Testcontainers setup. Triggers on phrases like "write a test", "fix this test", "improve coverage", "add an E2E case", "mock this dependency", or any edit to `*.Tests.csproj` or files under `tests/**`.

Use when writing, editing, or reviewing C# code that uses System.Reactive (Rx.NET); working with `IObservable`, `IObserver`, `Subject`, `BehaviorSubject`, `ReplaySubject`, or `AsyncSubject`; composing operators like `Select`, `Where`, `Merge`, `Switch`, `Publish`, `RefCount`, `Catch`, `Retry`, `ObserveOn`, or `SubscribeOn`; managing subscription lifecycle with `CompositeDisposable`, `SerialDisposable`, or `DisposeWith`; testing time-dependent reactive code with `TestScheduler` from `Microsoft.Reactive.Testing`; designing hot vs cold observables or enforcing the Rx serial-notification contract. Triggers on phrases like "observable stream", "reactive pipeline", "Rx.NET", "subject leak", "OnNext on wrong thread", "test scheduler", or edits touching files that import `System.Reactive`. Do NOT use for plain `IEnumerable` LINQ, `IAsyncEnumerable`, or TPL `Task`-based async.

Use when writing, editing, or reviewing ASP.NET Core Minimal API endpoints, route handlers, route groups, endpoint filters, or API responses; configuring OpenAPI document generation or build-time spec export; implementing endpoint validation or problem details error handling; using Server-Sent Events for streaming; writing integration tests against API endpoints; embedding Kestrel in a CLI tool. Triggers on phrases like "add an endpoint", "minimal API", "route group", "endpoint filter", "typed results", "OpenAPI spec", "Kestrel server", "API endpoint", "HTTP server endpoints". Do NOT use for MVC controllers, Razor Pages, Blazor Server, or SignalR hubs.

Use when writing, editing, or reviewing Mapperly (`Riok.Mapperly`) mapper classes; adding or changing a `[Mapper]`-attributed partial class; configuring `RequiredMappingStrategy`, `[MapProperty]`, `[MapperIgnore]`, or other Mapperly attributes; mapping between nullable Refit/OpenAPI DTOs and non-nullable domain records; debugging null-handling in generated mapping code (positional records, `required init`, initializer defaults); investigating Mapperly diagnostics such as RMG001, RMG020, or RMG089; inspecting source-generated output under `obj/Debug/*/generated/Riok.Mapperly/`. Triggers on phrases like "add a Mapperly mapper", "Mapperly null handling", "DTO to domain mapping", "generated mapper", or edits to files ending in `*Mapper.cs` under `ServarrApi/`. Do NOT use for hand-written mapping code or AutoMapper.

Use when weighing whether to extract a shared abstraction, base class, interface, or generic helper versus keeping code duplicated; evaluating DRY tradeoffs across parallel Sonarr and Radarr implementations, Refit or other code-generated API clients, OpenAPI/gRPC/GraphQL DTOs, or anti-corruption layers over distinct external systems; reviewing a PR that introduces a new shared abstraction spanning independent bounded contexts; considering unwinding a wrong abstraction back into duplicated copies. Triggers on phrases like "should I abstract this", "extract a base class", "these look duplicated", "DRY this up", "unify Sonarr and Radarr", "refactor to share", "rule of three", "wrong abstraction". Do NOT use for mechanical refactors within a single type or for trivial helper extraction inside one module.

Use when creating, editing, superseding, or reviewing Architecture Decision Records (ADRs) or Product Decision Records (PDRs) under `docs/decisions/` (including `docs/decisions/architecture/` and `docs/decisions/product/`); writing MADR-formatted decision documents; recording rationale for an architectural or upstream-driven choice; linking a decision to a TRaSH Guides upstream source; changing a decision's status (proposed, accepted, deprecated, superseded). Triggers on phrases like "write an ADR", "new PDR", "document this decision", "decision record", "add an architecture decision", or any edit to files matching `docs/decisions/**/*.md`. Do NOT use for general architecture documentation under `docs/architecture/`.

Use when adding, editing, or reorganizing entries in `CHANGELOG.md`; drafting release notes for Recyclarr; deciding which Keep a Changelog section (Added, Changed, Deprecated, Removed, Fixed, Security) a change belongs under; wording user-facing entries for `feat`, `fix`, `perf`, or breaking commits; phrasing deprecation notices or migration guidance; preparing the `[Unreleased]` section for cut. Triggers on phrases like "update the changelog", "add a changelog entry", "bump the changelog", "release notes", or any edit touching `CHANGELOG.md`. Do NOT use for commit message wording (see commit agent) or for internal `refactor`/`chore`/`test`/`ci`/`build` commits that do not require a changelog entry.

Use this skill when updating dependencies managed by uv: bumping a package version, upgrading the uv tool itself, updating torch/CUDA stack, switching transformers version, or regenerating the lockfile. Trigger: 'update dependency', 'bump version', 'upgrade uv', 'update torch', 'update lockfile', 'uv sync fails'.

Use this skill before committing ANY code change — this is a mandatory gate in the commit flow. Also trigger proactively when: you've made changes across multiple files and want to check consistency, you're unsure if a fix is safe, a change touches shared infrastructure (BaseTrainer, distributed, model loading, data pipeline), or a change is larger than a few lines. The review launches a subagent that checks implementation quality, multi-file consistency, and known constraint violations, then rates the change as safe/needs-attention/risky.

Use this skill for performance profiling and optimization. Two modes: (1) Analyze existing profile files (Chrome traces, memory snapshots) — write scripts to parse and summarize metrics per user requirements. (2) Generate profiles during development — configure ProfileConfig, run training, collect traces, analyze bottlenecks, and suggest optimizations. Trigger: 'profile', 'performance', 'slow', 'MFU', 'throughput', 'bottleneck', 'memory usage', 'trace', 'optimize training speed'.

Use this skill when adding a new optimized kernel or operator to veomni/ops/. Covers the full lifecycle: understanding VeOmni's ops architecture (KERNEL_REGISTRY + OpSlot dispatch, with a thin function-pointer shim for a few legacy global ops), implementing the kernel, registering it, adding tests, and documenting it. Trigger: 'add op', 'new kernel', 'add attention variant', 'new fused op', 'add triton kernel', 'optimize operator'.

Use this skill when adding support for a new model to VeOmni. Covers the full lifecycle: analyzing the HuggingFace model, creating model patches, defining parallel plans, writing configs, integrating with the trainer, and testing. Trigger: 'add model', 'support new model', 'integrate a model', 'new model support'.

Use this skill when adding or refreshing a patchgen-generated modeling file for a VeOmni model under its generated directory — GPU-only or GPU+NPU, dense or MoE, text-only / VLM / Omni-thinker+talker. Covers: creating GPU and NPU patchgen configs, using patchgen decorators (replace_class/override_method/replace_function/modify_init/add_post_import_block/drop_import_names), reusing sibling-model patches via name_map, handling MoE weight-loading (CheckpointTensorConverter + fused gate_up_proj layout), multimodal/VLM forward with Ulysses SP, excluding speech/vocoder subtrees in Omni models (talker/token2wav/DiT/BigVGAN), wiring __init__.py for the patchgen-generated classes, running codegen, and adding test cases. Trigger: 'port a model to patchgen', 'add patchgen for a model', 'transformers v5 migration', 'add NPU patchgen'. Do NOT edit files under generated/ manually — always regenerate via patchgen.

VeOmni-specific checklist for feature development and refactoring. Covers impact analysis across modalities, trainer hierarchy, data pipeline, and distributed code. Use before implementing any non-trivial change. For model-specific or ops-specific work, use veomni-new-model or veomni-new-op instead. Trigger: 'add feature', 'implement', 'refactor', 'reorganize', 'new capability'.

Use this skill for ANY bug, error, crash, wrong output, loss divergence, gradient explosion, test failure, CUDA error, distributed training hang, checkpoint load failure, or unexpected behavior. Covers both quick fixes (clear root cause) and complex debugging (unclear cause). Trigger: 'fix bug', 'fix error', 'broken', 'crash', 'doesn't work', 'fails with', 'loss NaN', 'training hangs', 'FSDP error', 'OOM'.

Create a pull request for the current branch. Handles uncommitted changes, generates a PR title matching the `[{modules}] {type}: {description}` format enforced by CI, and fills in the PR description template. Trigger: 'create pr', 'open pr', 'submit pr', 'make pr'.

Prepare journal manuscripts, conference papers, research posters, and grant documents using venue-specific formatting guidance and bundled LaTeX scaffolds. Use when selecting an official template, checking current page or anonymity rules, adapting academic writing to a venue, or inspecting a submission PDF.

Format and structurally validate local treatment-plan documentation after clinical decisions have already been supplied and verified by authorized licensed professionals. Use for source traceability, clinician-authored intervention records, goals and checkpoints, shared-decision records, reconciliation handoffs, and release gates—not for clinical decision-making.