Orchestrarium

Review keyboard, focus, semantics, contrast, and assistive tech.

Gate an approved phase or control-plane change: maintainability, cohesion, complexity, drift.

Advise on tradeoffs, ambiguity, cross-cutting concerns; never approve gates.

Run eligible review/QA via external provider; keep provenance.

Run eligible worker roles via external provider; keep provenance.

Coordinate approved delivery across stages, artifacts, gates, recovery, risks.

Review performance budgets, latency, throughput, memory, CPU, scale, cost.

Review auth, secrets, dependencies, data exposure, vulnerabilities.

Review usability, interaction clarity, accessibility, comprehension, flow.

Use when Claude needs evidence-backed manual visual verification of Windows desktop GUI before/after a fix across Qt, Avalonia, WinUI/WPF, WebView/native-child, or other Windows surfaces via screenshots, video frames, or live inspection.

Implement backend APIs, domain logic, persistence, integrations.

Implement data work: SQL, ETL, migrations, models, pipelines.

Implement web/client UI within design, accessibility, architecture constraints.

Implement geometry: transforms, intersections, meshing, indexing, predicates.

Implement 2D/3D graphics: rendering, shaders, scenes, materials, cameras.

Implement Qt model/view: models, proxies, delegates, selection, indexes.

Implement CI/CD, deployment config, runtime wiring, IaC, observability.

Verify phases with tests, regressions, edge cases, QA verdicts.

Implement Qt Widgets UI: dialogs, signals, focus, theme, high-DPI.

Implement toolchains, builds, CI graphs, packaging, caches, reproducibility.

Verify Qt UI: interactions, focus, high-DPI, themes, screenshots.

Implement scientific/data visuals: charts, plots, overlays, axes, legends.

Verify Windows desktop GUI (Qt/Avalonia/WinUI/WebView) with screenshots, video, or live inspection.

Use when Gemini needs evidence-backed manual visual verification of Windows desktop GUI before/after a fix across Qt, Avalonia, WinUI/WPF, WebView/native-child, or other Windows surfaces via screenshots, video frames, or live inspection.

Use when Qwen needs evidence-backed manual visual verification of Windows desktop GUI before/after a fix across Qt, Avalonia, WinUI/WPF, WebView/native-child, or other Windows surfaces via screenshots, video frames, or live inspection.

Adopt the Analyst role inline in THIS conversation for a trivial, bounded factual repository read. Investigate read-only and return a factual research memo with file and line references. Preserves current conversation context; does not claim isolation or independence. For a non-trivial or broad investigation, use the Agent tool with subagent_type: analyst instead.

Adopt the Architect role inline in THIS conversation for a quick-fix/fast-lane seam or blast-radius decision. Produce a design package from accepted research without writing implementation code. Preserves current conversation context; does not claim isolation or independence. For a non-trivial design in an isolated context, use the Agent tool with subagent_type: architect instead.

Adopt the Planner role inline in THIS conversation to upgrade the additive fast lane's inline plan into a phased delivery plan. Break an accepted design into small independent delivery phases with explicit file scope, dependencies, acceptance criteria, tests, checks, and quality gates. Preserves current conversation context; does not claim isolation or independence. For a non-trivial plan in an isolated context, use the Agent tool with subagent_type: planner instead.

Frame algorithms: invariants, assumptions, complexity, stability, correctness.

Investigate repo evidence read-only: files, symbols, flows, contracts, risks.

Design from accepted research: architecture, APIs, data, security, tests.

Frame physics/numerics: equations, units, discretization, solvers, validation.

Define performance budgets, bottlenecks, measurements, constraints.

Plan accepted design into phases, scope, dependencies, checks, gates.

Produce discovery briefs: users, constraints, metrics, scope, evidence.

Define SLOs, failure modes, resilience, observability, rollout, rollback.

Define threats, trust boundaries, controls, safe defaults, must-fix risks.

Design user flows, state behavior, hierarchy, usability criteria.

Adopt the Lead orchestration role in THIS conversation (never spawned as a subagent) — classify approved work, route it to leaf specialist subagents via the Agent tool, gate artifacts, own work-items recovery and integration. Use when a team template marks requiresLead true, when the operator invokes /lead, or when resuming a lead-managed work-item.

Adopt the Product Manager role inline in THIS conversation for quick intake/scope framing when priority is unclear. Own roadmap decisions and admission into discovery or delivery. Preserves current conversation context; does not claim isolation or independence. A formal cross-initiative roadmap decision, or admitting work that will gate other work, must route to the Agent-tool subagent (subagent_type: product-manager) instead, to keep admission authority separate from the executor.