Load local session state + compiled truth. Optionally pull remote truth first.

Offline-first: always returns local data even if remote is unreachable or disabled.

Update this agent's focus and warn on overlaps with other non-stale agents.

Write a durable fact remotely if online; otherwise queue locally for later flush.

Flush the offline queue to remote, optionally consolidate + pull compiled truth.

Pull compiled-truth markdown from the remote host into the local cache.

Publish an event to the AgentRelay Event Bus.

Poll events from the Event Bus since sequence number since_seq.

Subscribe an agent to specific event types on the Event Bus.

Delegate a task to a headless CLI agent or role.

The default agent='auto' selects an installed worker by capability. Pass required_capabilities when the orchestrator already knows what the task needs, and exclude_agents to keep the human-facing orchestrator out of the worker pool. Direct agent selection and static roles remain available as explicit overrides. If worktree is True, the agent runs in an isolated git worktree branching from cwd. checks are shell commands run after the agent finishes (for example a test command); read their outcome with job_review before spending anything on the job's output. Optional memory_project enables automatic local session-memory bootstrap and finalization for the given project key without storing raw prompts in memory.

Preview automatic worker selection without launching a job.

Read automatic delegation policy. Call when deciding whether to delegate or offer it.

List dispatch agents with live binary availability and routing capabilities.

Check the status of a delegated job.

Wait for a background job to finish and return its review as a completion ping.

Call this immediately after delegate_task(background=True) instead of repeatedly polling job_status or ending the turn. The pending MCP call resumes when the job completes, fails, or is cancelled, so the orchestrator can review and report the outcome without the human babysitting it.

Get a mechanical review verdict for a job, including check results and diff summary. Call this before job_result to skip reading the output of work that did not pass.

Fetch the result file content for a completed job.

Cancel a running job and kill its process tree.

List available models for the given agent or all agents. Use this for choosing a model before delegating a task.

List configured roles and their agent, model, and effort mappings.

Prefer specifying a role over an agent name when delegating tasks so that model and effort configurations are applied automatically.

Report local paths, queue depth, and remote reachability.

Start a tracked local-memory session and return its session ID.

Save validated, best-effort-redacted structured session memory.

Mark a memory session as completed or failed.

Retrieve bounded project context, prioritizing open and recent sessions.