team-assemble
Assemble a task-fit Agent Team dynamically, request approval, then execute the staged lifecycle.
Purpose
Use this skill when fixed engineering-only teams are not enough and the roster should adapt to task shape.
Trigger
- User asks to "do it as a team"
- Task mixes domains (for example research + implementation + reporting)
- Team composition and model budget must be explicit before execution
Workflow
- Analyze task shape (domain, output target, risk posture).
- Select domain specialists and format specialists separately.
- Spawn parallel research lanes (
omi-researcherxN) when breadth is high. - Assign decision lane (
omi-consultantoromi-architect) and quality lanes (omi-reviewer,omi-verifier,omi-debugger). - Assign model profile by role:
- judgment/gating:
glm-5 - execution-heavy:
glm-4.7 - broad low-risk exploration:
glm-4.7
- judgment/gating:
- Ask approval gate: "Proceed with this team? (yes/no)".
- On approval, run the staged pipeline (
team-plan -> team-prd -> team-exec -> team-verify -> team-fix) with explicit handoffs.
Output Template
## Team Fit Analysis
- ...
## Proposed Team
| Role | Agent | Why | Model | Lanes |
| --- | --- | --- | --- | --- |
## Collaboration Protocol
- ...
## Approval Gate
- approval detected:
- prompt:
## Execution Summary (only when approved)
- ...
Notes
- Keep role assignment minimal and explain why each role exists.
- Prefer explicit verify/fix triggers over vague quality claims.