/as:studio

Before acting, read the host contract, this component's declaration (skill:studio), and applicable workspace semantics. Load only the required modes from the shared catalog.

<!-- architecture-studio:harness-compatibility -->

Use the host adapter for native invocation. Apply the native workspace operations with the host’s own tools and ordinary task-specific code. Operation names identify semantic scope, not function calls or a dispatcher. No Arch Studio runner, executable download or reconstructed helper is required. Resolve canonical context before existing-workspace edits; fresh setup validates its explicit target and ancestor boundary instead. Existing exact authorization persists; ask only for material missing information or missing permission.

Own STUDIO.md, project registration, synchronized registered Status and the studio-only connector boundary. Read moments for state-aware welcome and no-folder behavior. Do not repeat a declined setup nudge. A supplied task proceeds directly; no welcome/menu replay. Tools-only and valid standalone project use require no studio scaffolding.

Before changing records

Before persistent changes, follow the native mutation sequence: Inspect → Prepare → Verify preparation → Apply → Verify result → Complete. Inspect pending work first. Finish and verify preparation before the first canonical write: separately reread every saved original and prepared file, check its complete actual bytes, establish durable saving, and inspect actual permissions, ownership and ACLs. Merely writing recovery files does not finish verification. Apply only that verified set. Mark complete only after fresh readback validates the entire affected result and its relationships. Read-only requests and inline drafts need no publication sequence.

Fresh setup prepares the whole declared studio tree and identities. Project-mode registration changes STUDIO.md while guarding the existing project's identity; portfolio registration also includes any permitted removal of a header-only local TASKS.csv and the project task-pointer update in the same prepared set. Status/archive changes prepare both STUDIO.md and PROJECT.md before either publisher runs. Task-mode changes include every affected register and ownership reference. Single-setting edits remain bounded to their declared file/field; optional preference/connector edits retain their separate exact authorization.

Setup

Read the studio policy contract for fresh setup or an explicitly requested policy configuration. Ask once whether the user has an AI-use policy or wants the offered default, show the default before adoption, and allow decline. Existing references stay in place; a default is created only at the resolved Standards root's governance/ai-policy.md. Record the reference and adoption choice in STUDIO.md's Data governance section. Include that exact file set in the existing setup preview and readback. Preserve already-configured choices on unrelated edits; Skill Maker only consumes them. No policy is silently adopted or migrated.

Use studio.init for a confirmed exact fresh target. Gather only missing studio name, working units, jurisdiction (No default is valid), naming policy (as, firm, none), folder taxonomy/managed roots, document path template and firm stage vocabulary keyed by project Kind. Naming and taxonomy are separate advisory choices. Preserve confirmed human names/casing. The shipped building/software/initiative stages are Arch Studio-owned defaults, not an asserted external standard.

Show the exact target, files, managed roots and data boundary: records stay in the user workspace; content supplied to a configured model follows that provider's terms; setup creates no ALPA account/cloud storage/Git repository. One concrete setup gate covers those choices. Format version stays 3; Document model 1 identifies the fresh record shape. No existing workspace conversion or backward-compatibility setup is provided.

Finish setup preferences with: “Would you like this assistant to use Arch Studio by default for architecture and project work?” Follow the preference protocol. When the exact scoped preference edit has been previewed, the same setup approval can cover it; a general yes never approves an unseen file edit. Otherwise retain session guidance and obtain only missing concrete authorization. Do not persist a global preference implicitly.

MCP instruction delivery does not install lifecycle hooks. Background update preference is an installed-package feature outside this MCP workflow; report it unavailable here without inspecting or creating package state. Do not infer enablement from connection or general setup approval.

Apply the reviewed studio.init preparation with native tools. Reopen STUDIO.md, identity, connector manifest (mcpServers empty), managed roots and host skill roots. Installation/connection checks do not prove workspace creation.

Projects and settings

• studio.status lists valid and invalid registrations, identity drift and task mode read-only. Unknown or archived status does not hide a project. No authentication/network checks are implied.
• Create-project orchestrates project.init then studio.register; each uses the same already-approved identity/target. A partially completed project is reported and registration resumed without recreating it.
• studio.register requires an existing new-model descendant with immutable unique Project ID and Folder ID. Paths are relative cached locations; do not infer identity from a human folder name.
• studio.set-status and studio.archive change only the selected registry Status cell and the matching PROJECT.md Status Value cell. Preserve its existing Source and Date cells exactly unless the user separately authorized those specific provenance changes; do not replace Source with the operation name. Before publication and on actual readback, compare the field-level diff across both files and reject any other changed cell. Activity never establishes completion or acceptance.
• studio.set-naming changes the advisory policy/convention without renaming existing IDs or folders.
• studio.task-mode switches only empty current-model task registers between project and studio portfolio ownership, preserving the one-writer topology. Populated registers require a separately specified current-data operation, not an automatic merge/split or historical converter.

Only studio owns .mcp.json, at studio root. Setup leaves it empty; connector configuration/authentication requires separately available capabilities and scope. Projects never receive this manifest. Firm extensions stay in the relevant host's studio skill root.

General domain work may hand off once to Norma with scope, completed outputs and existing authorization; an explicit studio administration task stays here. Main-harness agent creation follows approved roles/assignments and real capabilities. Do not claim that instructions or a welcome executed a workflow.

Optional preferences and connectors

Use studio.preference only for the optional default-assistant choice, applying the preference protocol with the verified native instruction mechanism and concrete authorized diff. No persistence claim follows from connection or a saved file alone. MCP background lifecycle update checks remain unavailable; do not invoke installed-package update operations through this workflow.

studio.connectors owns the studio .mcp.json boundary. Fresh setup creates exactly an empty mcpServers object. A later explicitly requested configuration must use verified connector/host documentation and a reviewed exact change, preserving other entries. Configure only the known authorized server; authentication, external messages and account provisioning require their actual separate capabilities and authorization. No arbitrary connector endpoint or secret is inferred from workspace setup.

After administration, use the shared completion reporting contract. Report actual changes and readbacks; a connection or welcome is not workspace creation.