ALWAYS CALL THIS FIRST, before any other komuta tool. Classify the user's request into exactly one intent and pass it here. Returns the workflow contract: a FIXED ordered step list (never skip, reorder, or invent steps) and the EXACT tool set allowed for the task — no other komuta tool may be called while this workflow is active. Call again with a new intent when the user's request changes. Intents: MANAGE_JOBS — Run, cancel, suspend or resume existing Jobs/CronJobs, or inspect their configuration and history. New job creation uses DEPLOY_STACK. | MANAGE_SERVICE_RUNTIME — Inspect service configuration, ports, health probes, routing, dependencies; change fixed replicas, sleep/wake, automatic sleep (turn idle sleep on or off), maintenance mode (turn it on or off, change the visitors' maintenance message) or control a rollout. Also the organization's monthly free awake hours and a free-plan service's keep awake (always on) add-on: its status, price, purchase or turning it off. | ACCOUNT_BILLING — Account-wide costs, statements, build allowance, wallet, payment links, approved wallet payments, the Doktor AI add-on subscription and asking Komuta for a white-label report or evidence pack pilot. | MANAGE_ALERTS — Inspect alert rules, templates, notification channels and history; deploy, undeploy or enable/disable existing rules and channels; create an email notification channel or send a test notification through a channel. Also scheduled Service Doctor health reports: list, create, change, pause, delete or preview when and where they are sent, or send one right away. | SECURITY_AUDIT — Inspect security audit history and evidence for an organization or service. | MANAGE_ADDON_LIFECYCLE — Manage an existing Postgres, Valkey or RabbitMQ instance: backup, restore, restart, plan change, suspension, deletion scheduling or credential rotation. | MANAGE_API_GATEWAYS — Create, inspect and manage API gateways end to end: settings, connected APIs, endpoints and their policies (cache, limits, authentication, header and path rewrites, request checks), callers' keys and usage plans, OpenAPI import, test requests, configuration history, export/import, applying changes, scaling, bandwidth and version. | PLATFORM_ADMIN — Platform administrator: inspect cross-tenant services, platform capacity, addons, tenant status and platform health; manage tenant suspension, limits and maintenance. | ADMIN_TENANT_BILLING — Platform administrator: inspect or change another tenant's billing, spending limit, wallet balance or statements. | ADMIN_BACKGROUND_JOBS — Platform administrator: inspect workers and background/recurring jobs, retry jobs or delete failed work. | DEPLOY_STACK — ANY request to publish/provision/deploy a new app, Git repository, Service, Job or CronJob goes directly to this workflow, even when the user never says Stack or YAML and even for a single service. Also use it to create/use komuta.yaml, deploy multiple services/addons/domains together, validate/plan/apply a Stack, resume a Stack run, inspect Stack drift, or export an applied Stack. Only a redeploy of a workload that already exists on Komuta remains REDEPLOY_SERVICE. | REDEPLOY_SERVICE — User wants to redeploy a workload that already exists on Komuta: deploy a new version, re-run/re-trigger its pipeline, or change the branch it deploys from. NOT for publishing a brand-new workload (use DEPLOY_STACK). | MANAGE_AUTOSCALING — User wants to see, enable, disable, or change a service's autoscaling / HPA: min or max replicas, target CPU %, target memory %, or the scale-up/scale-down stabilization windows. | MANAGE_AUTO_DEPLOY — User wants to see, enable, disable, or change a service's auto-deploy / deploy-on-push behavior: turn it on or off, include/exclude path patterns, the cooldown between triggers, or commit-message skip directives. NOT for manually redeploying right now (use REDEPLOY_SERVICE). | MANAGE_ENV_VARS — User wants to see, add, or change environment variables / runtime configuration / secrets of an existing service — including wiring a managed addon's connection string into a service, copying stored secrets from another service, or importing an environment group's variables into a service. NOT for creating, filling or pushing environment groups themselves (MANAGE_ENV_GROUPS), for creating the addon itself (ADDONS) or for a first deployment (DEPLOY_STACK handles initial configuration and secret handoff). | MANAGE_ENV_GROUPS — User wants to see, create or fill an environment group (a shared set of variables used by several services), see which services use it or have fallen behind it, or push its current values to those services. NOT for one service's own variables or importing a group into one service (MANAGE_ENV_VARS). | MANAGE_DOMAINS — User wants to add/bind a custom domain (their own hostname) to an existing service, check its DNS/TLS verification status, re-check after they've updated DNS records, or remove a domain. NOT the auto-generated *.komuta.app hostname shown by komuta_get_service_overview — that one isn't managed here. A domain requested as part of a first deployment belongs in DEPLOY_STACK. | MANAGE_STORAGE — User wants to see, attach, resize/upgrade/downgrade the plan of, or remove a persistent storage volume (disk/PVC) mounted into a service's container. NOT a managed database/cache/queue's own storage (that's ADDONS) and NOT compute sizing (that's MANAGE_AUTOSCALING). | MANAGE_ACCESS_PROTECTION — User wants to see or change who may open an existing service from the internet: turning access protection on or off, allowed IP addresses, requiring Komuta sign-in, path rules (blocked pages, webhook paths open without sign-in, pages for chosen people only), an expiry, telling the application who signed in and how to verify it, sharing the service with members, the organization, linked organizations, people by e-mail, everyone at a domain or everyone who signs in with the organization's identity provider (optionally some of its groups), the organization's identity providers for visitors, service tokens for machines, which other services may reach it over the private network, the access log of sign-ins, page views and refusals, whether a particular person, visitor or machine can open a given page and why, or the organization-wide setting for sharing outside the organization. | ADDONS — User wants a database, message queue, or cache — create one, list them, or get connection info. | STATUS_OVERVIEW — User asks what they have running, whether a service is comprehensively healthy or at risk, the state of a service or addon, a general account overview, wants to see/search a service's logs or request traces (failing/slow endpoints), or asks about a service's billing — current charges, cost so far, or its invoice history. | MANAGE_RECOMMENDATIONS — User accepts, defers or dismisses a critical recommendation Komuta returned for a service, wants its action plan prepared, approved, marked as done, or verified, or asks Komuta to fix a Service Doctor finding itself (one-click fix), or wants a Service Doctor README badge turned on, renewed or turned off. NOT for asking whether a service is healthy (STATUS_OVERVIEW). | GENERAL_CHAT — No platform action requested — questions, chit-chat, or explanations. Includes 'what is Komuta?', 'what can it do?', how deploys/addons/pricing work conceptually, and 'how do I get started / get an API key?'. Returns: { workflow, title, steps, allowedTools, recovery, rules }. Show user: nothing verbatim — steps/allowedTools/rules are agent-facing orchestration instructions; act on them and speak to the user in your own words. Never show user: the raw steps/allowedTools/recovery/rules payload — never dump the workflow contract into the chat. Sends: nothing to the backend — only switches this session's active workflow (overwriting the previous one's context).