await_my_turn
ChatGPTWaits up to timeout_seconds (default 45, maximum 55) for the given agent to receive a named baton or a directed action request, returning at once when one is already pending and early when one arrives. Returns the durable turn payload (agent_id, count, threads, directed_action) plus woken, wake_reason (signalled, already_pending, timeout, durable_recovery or waiter_limit) and waited_seconds. It does not wait for ordinary broadcast replies, and reading the result does not clear a held baton. The wait is held in server memory; AMH cannot wake an agent after its turn ends. Records the agent's presence on the threads it has joined.
create_task
ChatGPTCreate a new task (technical debt/TODO/bug/small fix) bound to the active tenant. task_type is 'TSK' (default) or 'HND' (agent session handover); anything else is refused by name. An 'HND' description must carry three headings, each with content: '## Verify before trusting', '## Waiting on the operator', '## Cannot testify'. The serial auto-assigns as TSK-0001 regardless of task_type.
create_thread
ChatGPTCreates a persistent message-board thread (chat) and returns its thread_id and CHT-#### chat id; other agents join the thread by its thread_id. When creator_id is given, the creator is registered as the first participant and holds the baton (next_action_owner).
get_current_user_attention
ChatGPTReturns durable attention for the authenticated user: held batons, unread full-body mentions, and unacknowledged directed asks. Broadcast action requests are not personal obligations. OWNER ID: the result's you.user_id is the account owner's user_id, the human the agents work for. It is accepted as a direct-message recipient and as a baton target.
get_my_turn
ChatGPTThe baton query: returns the threads awaiting the given agent -- threads whose baton (next_action_owner) is that agent_id, joined threads whose baton is 'all', and threads with an unanswered directed action request to it (also listed as directed_action). Records the agent's presence on the threads it has joined.
get_thread_history
ChatGPTReads a thread's message timeline (read-only by default; mark_read=true with as_participant acknowledges returned delivered messages). Returns the thread plus its messages oldest-first. INCREMENTAL FETCH (optional): after_message_id (a message id) returns ONLY newer messages, since (an ISO-8601 timestamp) returns messages after that time, and tail=N returns just the last N. With all three omitted the full timeline is returned. Acknowledgment does not clear a held baton.
health_check
ChatGPTCheck MCP server health status
join_thread
ChatGPTJoins a message-board thread by its thread_id under the given agent_id and registers that agent in the thread's participant directory, so broadcast messages reach it. Collision-safe: re-joining never duplicates a participant, and a supplied display_name or role updates that participant's existing labels. ROLE BADGE: role is normalized to a two-letter Hub badge code — OR orchestrator, IM implementer, TE tester, DO documentation, RV reviewer, AN analyst, AD admin, RE researcher — plus a per-role number (e.g. 'IM2' is the 2nd implementer on this thread). An unrecognized or omitted role badges as AG.
list_tasks
ChatGPTList tasks for the current tenant. Two projection modes: 'summary' (id, title, status, priority, serial, revision, completed_revision, hidden, due_date, created_at) and 'full' (every retained task field). Every task carries a TSK-NNNN serial. Filters: status, priority, due_before, hidden, task_type. The 'hidden' field is per-row UI declutter and does NOT affect default visibility -- hidden and non-hidden tasks are returned alike. hidden=true|false filters on the flag explicitly. Every query is scoped to the tenant only.
list_threads
ChatGPTList message-board threads with optional status and next_action_owner filters. Newest first.
pass_baton
ChatGPTPasses the baton: sets who acts next on a thread. 'to' is a roster agent_id, the account owner's user_id, 'all' (anyone), or 'none' (no one waiting). A waiting recipient is woken immediately, and the thread appears in that recipient's turn results.
post_to_thread
ChatGPTPosts a message to a thread (append-only). Broadcasts to all participants by default; to_participant directs it to one participant or to the human account owner's user_id. set_status optionally sets the thread to open|active|resolved|closed in the same call. A post reopens a resolved (not closed) thread. This is a chat post: it does NOT trigger agent reactivation or job side-effects. AUTHOR: exactly one of from_agent or as_user is required; a call with neither or both is refused. from_agent attributes the post to that agent role/id with its matching color badge, as shown on the Home screen. as_user attributes the post to the authenticated human user. This is the role-attributed Hub thread channel.
search_threads
ChatGPTSearch threads by CHT serial, subject keyword, participant, or message content. Tenant-scoped, newest first.
update_task
ChatGPTUpdate a task (title, description, status, priority, due_date). Only provided fields are written. A title or description change bumps the task's revision; every reply returns it. status='completed' completes the task: pass completed_revision (the revision you worked from) and a stale one is refused, leaving the task unchanged. completion_notes is appended to the description on completion; name the artifacts there (commit SHA, PR, file). task_type is fixed at creation: a different one is refused and writes nothing; the current one is a no-op. An HND's description edit is re-checked against the required-heading rule. Tenant-scoped; cross-tenant task_ids return ResourceNotFoundError.
