Command Center AI Documentation

The package owns the documentation of what it owns. Decisions and guides move here in the same step as the code they describe (ADR 096, section 7).

Decisions

• ADR 060: Session-Backed Chat Requests
• ADR 087: Queued Chat Messages While the Agent Works
• ADR 090: Agent Icons In Command Center Surfaces
• ADR 093: Client-Verified Agent Readiness
• ADR 096: The Chat as One Independent Package
• ADR 098: One Communication Contract for Every Agent
• ADR 099: Local Agents Through the Same Engine

These records moved here from Command Center and keep their numbers and titles, because the code cites them. New decisions about the package are SDK ADR records in adr/ with Package: @dev-mainsequence/command-center-ai, numbered in the repository's one SDK ADR sequence (SDK ADR 012).

Guides

• Getting started: install the package next to the SDK, its agent skills, and upgrades.
• Build an AI application: the decisions first, then the stylesheets, the engine, the thread, and the settings, verified on the stand-in.
• The right rail and the expanded rail: where the conversation sits in the application, and how it looks.
• Connect to the platform: the connection, the request-URL rewrite and the forwarder, tokens, and the failure states.
• An Agent on your machine: the same chat over an Agent run with ms-tau in local mode, through the dev server's localAgentProxy().
• Model providers: built-in provider sign-in, Organization custom providers, and the test conversation.
• The conversation contract: what the package sends an Agent, what the thread draws of the answer, and what a reload keeps.
• Backend connection: the connection input, the clients, and the runtime contract.
• Session engine: the engine's inputs, the default session, its context and hooks, the model catalog store, and browser storage.
• Thread UI: the thread and its inputs, the stylesheet, readiness in the thread, and the message actions.
• Model providers: the settings screens, their state, and the rules for credentials, custom providers, and the test conversation.
• AgentSession resolution: how the engine chooses, hydrates, and connects a session and enables sending, the same for every Agent, and the sessions an application lists and manages.
• Provider errors

Task and agent-skill map

The package ships one agent skill per task below, and each has a human guide with the same public entrypoints and rules. tests/cli/docs-skills.node.mjs fails when a packaged skill is missing from this table or its guide does not exist. The skills sit in lanes, as the SDK's do.

Task	Human guide	Agent skill
Install the package next to the SDK, refresh its skills, upgrade, and choose the focused skill	Getting started	general/use-command-center-ai
Decide where AI lives in the application and which pieces it needs	Build an AI application	general/build-command-center-ai-application
Mount the engine, the thread, and the settings, and verify on the stand-in	Build an AI application	engine/mount-agent-conversation
Compose the right rail and the expanded rail	The right rail and the expanded rail	ui/compose-command-center-ai-rail
Open, list, archive, and link Agent sessions	AgentSession resolution	sessions/manage-agent-sessions
Reach the platform and the Agent runtime from the application's origin	Connect to the platform	backend/connect-command-center-ai-to-the-platform
Connect providers, administer custom providers, and test a custom model	Model providers	model-providers/manage-model-providers
Design an Agent's answers for what the thread shows	The conversation contract	contracts/design-agent-conversation-capabilities

The skills install into .agents/skills/command-center-ai/ of the repository that installs the package; see the agent scaffold. The Command Center SDK's general skills send an agent here when an SDK application needs a chat or AI capabilities.