orchestrator

Screenshot-first visual end-to-end validation policy for browser-rendered UI and webview jobs after verifier approval.

Default engineering workflow. Classify a technical request, route it through the native T1/T2/T3 engineering fleet, and return only independently verified completion to the orchestrator.

Safely operate Claude Code's opt-in advanced capabilities—sessions, worktrees, agent teams, Skills, MCP, plugins, hooks, Channels, schedules, goals, and Agent SDK integrations—when the user explicitly asks for one. Do not use for an ordinary build request.

Routes selective local Codex plan, code, verification, and visual judgment with evidence-backed blocking and strict cost caps.

Shared evidence, accessibility, context, and handoff rules for the local design-agent fleet.

Produces contextual creative theses and bounded visual directions without generic model defaults.

Creates one disposable selected-direction prototype under the isolated design prototype root.

Independently reviews design packets for evidence, coherence, accessibility, completeness, specificity, and engineering readiness.

Native Claude Code routing and handoff policy for the T1/T2/T3 engineering fleet and Engineering Lead. Use for an engineering job after planning, an escalation, or an explicitly authorized extreme advisory team.

Defines conditional motion hierarchy and code-native asset direction with accessible fallbacks.

Mandatory authority, routing, direct-handoff, approval, and context policy for the read-only Planner agent. Use for every PLAN_JOB and team-based planning task.

Add proportionate risk, verification, rollout, rollback, security, accessibility, and stale-plan controls to a plan. Use for medium/high-risk, migration, integration, browser UI, or security-sensitive work.

Mandatory evidence-first policy for Researcher. Use for every research request to govern intake, prior-evidence reuse, profile/mode selection, source-ledger evidence packets, authority, and direct Scribe handoffs.

Mandatory evidence, context, handoff, and editorial policy for Scribe. Use for every DOCUMENT_JOB and every team-based document handoff.

Write plans, requirements, instructions, ADRs, and decision records that preserve authority, scope, assumptions, and acceptance checks.

Mandatory production-first execution policy for t1-engineer. Use for every bounded implementation job to preserve scope, select task profiles, enforce full quality review, and return evidence.

Preserve evidence and ownership when T2 takes over an unresolved T1 engineering job.

Specifies journeys, information architecture, responsive behavior, content hierarchy, and accessible interaction states.

Evidence-first independent verification policy for the post-T1 verifier. Use for every VERIFICATION_JOB.

Defines an implementable visual language and DTCG-aligned primitive, semantic, and component token system.

Define or revise a documented reusable engineering mode using native Claude Code Skills and custom agents. It does not create JavaScript workflow runtimes.

Analyze an unfamiliar, multi-file, or cross-boundary repository task before planning. Use to map affected paths, local conventions, interfaces, tests, dependencies, and no-touch boundaries.

Turn a non-basic request into a bounded, executable specification and ordered work plan. Use for multi-step features, documents, integrations, migrations, architecture, and design-uncertain tasks.

Research browser and webview UI safety, accessibility, and automation constraints, including the mandatory prohibition on native browser dialogs.

Write README, API, architecture, operator, and developer documentation from verified repository and engineering evidence.

T1 UI wiring and verification playbook. Use for frontend behavior, forms, flows, state transitions, responsive layouts, or accessibility-sensitive changes.

Force an evidence-first research pass before advice or an engineering handoff.

Review or rewrite an existing non-code document for clarity, consistency, accessibility, and the project's anti-tell writing rules.

Turn a Researcher RESEARCH_EVIDENCE packet into a cited report or research briefing without changing the evidence or its authority boundaries.

Deep codebase and architecture research playbook. Use for architecture, repository, system-design, migration, or implementation-blueprint research.

Evidence-backed comparative research playbook. Use when the decision concerns two or more options, vendors, architectures, frameworks, or approaches.

Informational high-stakes research guardrail. Use for legal, medical, financial, safety, privacy, security, regulated, or materially consequential topics.

Evidence-driven root-cause research playbook. Use for incidents, defects, regressions, unexplained behavior, or failure investigation.

Research source-quality and evidence-audit playbook. Use for deep, current/volatile, comparative, high-stakes, or disputed research.

Deep strategic or general research playbook. Use for broad market, product, business, policy, or multi-domain research that needs synthesis and recommendations.

T1 API and integration contract playbook. Use for HTTP, RPC, events, queues, database boundaries, authentication, or external-service changes.

T1 measurement-driven performance optimization playbook. Use when a job claims latency, throughput, rendering, query, memory, or resource concerns.

T1 behavior-preserving refactor playbook. Use when reorganizing, simplifying, or reducing duplication without an approved behavior change.

T1 codebase route-tracing playbook. Use for unfamiliar, cross-file, stateful, or side-effecting implementation work before changing code.

T1 security, reliability, and recovery playbook. Use for changes that cross trust, data, auth, secrets, persistence, or critical failure boundaries.