CloudClawer/CloudClawerDocs
BlogSign InSign In
Warp SpeedRoles & Orchestration
By Feature · Roles & Orchestration

Roles & Orchestration

Promote your workflows into reusable roles, run as foreground subagents under the CoA.

Introduced at Warp 3. You stop running one agent and start running a small team — each a specialized role with its own workflows. The orchestration is deliberately simple: foreground subagents, launched and collected by the Chief of Agents, with no back-channel between them.

Role files for multi-agent work

Consolidate the workflows and checklists you built at Warp 2 into roles. Different roles get different workflows — an Engineer differs from a blog writer; a Growth Engineer does outcome engineering, code and content that leads to sales. A role file is just durable, reviewable instruction, like everything else the forward-deployed agent writes.

Consolidate Warp 2 workflows/checklists into role files (Engineer, blog writer, Growth Engineer, …).

Foreground subagents under the CoA

Roles run as foreground subagents launched by the Chief of Agents. There is no inter-agent messaging — each subagent runs to completion and returns its report. That means you must front-load each subagent's full instructions: you cannot message it mid-run to course-correct.

Launch roles as foreground subagents via the Chief of Agents.
Front-load every instruction — no messaging once a subagent is running.

Checklists & design principles

Checklists enforced by roles.(Strict mechanical checklists aren't implemented yet — enforcement is by convention.)
Design principles — if you have them written down, commit them so every role inherits them.
Notice the shape: roles are Warp 2's workflows promoted into reusable, specialized form. You're not granting new autonomy here — you're building the organization that Warp 4 will eventually run on its own.
© NeuralAccel 2026