Collaborate through team.md

Turn a multi-Agent goal into one explicit protocol document. teamhub room collaborate publishes it as this Room's team.md protocol and notifies every declared participant except the publishing Agent, including a separately appointed Orchestrator. Members read the current protocol with teamhub room protocol; no local team.md projection is mounted.

Read Collaboration Patterns before choosing a pattern.

1. Confirm teamMD is the right tool

Publish a team.md only when all of these are true:

• the work genuinely needs more than one Agent;
• every Agent you intend to name is already a member of this Room;
• you can state one shared goal and each member's responsibility.

Do the work yourself when a single Agent suffices. Follow your Context's permissions for other coordination tools; this Skill does not authorize creating a Task or Workflow.

If an Agent you need is not in the Room, add it first. teamhub room collaborate rejects the whole document when any declared Actor is not an active Room member; it does not add anyone.

2. Choose one pattern

The value must match one of these exactly. TeamHub does not normalize spelling, spacing, or case.

3. Choose the Orchestrator and write the document

Assess the capabilities of the Agents already in this Room and choose one suited to coordinating the user's request. You may publish the plan with another Agent as Orchestrator and yourself as a Team Member. If no other Agent is suited to orchestrate, act as both Orchestrator and contributor. Describe your coordination and concrete work responsibilities in the Orchestrator's Role; declare your Actor ID only once, without repeating yourself under Team Member.

---
collaborate_pattern: fan-out-and-synthesize
---

## Task Goal

Summarize the competitive landscape for workflow platforms.

## Orchestrator

- Coordinator
  - ID: actor_agent_01

## Team Member

- Researcher
  - ID: actor_agent_02
  - Role: Collect raw vendor material.
  - Report to: Coordinator
  - Receive from: Coordinator

- Synthesizer
  - ID: actor_agent_03
  - Role: Merge findings into one brief.
  - Report to: Coordinator
  - Receive from: Researcher

Rules the document must satisfy:

• the frontmatter holds exactly one key, collaborate_pattern, with a value from the table above;
• ## Task Goal is present and non-empty;
• ## Orchestrator names exactly one suitable Room Agent, with an ID; it may be you or another Agent;
• ## Team Member names at least one Agent; every name and every ID is unique;
• every ID is an Actor ID of an active member of this Room, not a display name;
• Report to and Receive from reference names declared in this same document, or none. Separate several names with commas;
• Role is optional but makes the protocol usable; write what that member owns, not how it works.

Name each participant with a short role-like name (Researcher, Reviewer). Relationships reference these names, not Actor IDs.

4. Publish and hand off

Save the complete document to a UTF-8 file and run teamhub room collaborate --file <plan-path> --operation-id <unique-plan-id> --json. It publishes in the configured Room. No source Message ID is required; publication is a Room collaboration event. On success it returns created: true, room_id, revision, and notified_actor_ids — the declared participants TeamHub notified. The publishing Agent is excluded. A separately appointed Orchestrator is included and takes over coordination; follow its assignments while completing your own contributor responsibilities.

Read current participants with teamhub room members; use its real Actor IDs. Publishing replaces any previous team.md for this Room and bumps revision. There is no partial update: send the whole document every time.

Publishing notifies the declared participants to read the protocol. Entry Agents can begin work whose inputs are ready; other members wait for an upstream handoff. The designated Orchestrator handles any additional assignments. Do not send duplicate start messages for work already launched by the publication notification.

Report to and Receive from are declarations that Agents and Humans read. TeamHub does not enforce them at run time: it will not block a message that ignores the topology. Follow the protocol you published, and say so plainly when you deviate.

5. Handle failures precisely

A rejected submission writes nothing and notifies nobody. The error carries one issue per broken rule, each with a path, a code, and a message:

• team_md_pattern_invalid / team_md_pattern_missing: use one of the five exact values;
• team_md_actor_not_room_member: that Actor is not in this Room. Add it to the Room or drop it from the document; do not swap in a different Agent to make the call succeed;
• team_md_reference_unknown: Report to or Receive from names someone the document never declares. Fix the name or declare the member;
• team_md_actor_id_missing / team_md_actor_id_duplicated / team_md_name_duplicated: each participant needs one unique name and one unique ID;
• team_md_section_missing / team_md_section_unknown: the document has exactly three sections — Task Goal, Orchestrator, Team Member.

Fix every reported issue and resubmit the whole document. Do not retry the identical content, and do not report that a protocol was published after an error.