Orchestration

Turn a repeated task into a budgeted, checked workflow file with audit traces

Split project work into dependent tasks, run parallel agents and ship pull requests

Run four parallel C-suite agent reviews and a day of autonomous business work

Pick the model and effort per workflow stage and confirm estimated cost first

Send a request to another agent in a CCB multi-agent workspace

Send a result to a reviewer and loop through a bounded pass or rework cycle

Carry out one scoped coding or investigation task and return evidence

Diagnose stuck replies, busy agents and lost callbacks in CCB multi-agent runs

Diagnose one named CCB agent from runtime logs, job lineage and provider data

Diagnose and repair CCB job, reply and callback chains for stuck work

Check CCB daemons, tmux panes, queues and replies to find what is broken

Restart, remount or reload stuck CCB agents and provider contexts safely

Batch planner questions into one short clarification set and map answers back

Reference for the start and stop lifecycle of dynamically spawned agents

Turn a chat with the user into workflow requests and curated final artifacts

Verify a CCB worker node and reject hidden fallbacks or missing evidence

Pick a route and propose a one to four workgroup bundle for multi-agent work

Reference for requesting capacity for orchestrator loops in multi-agent runs

Check planner task packets for readiness without changing authoritative state

Draft a planner backfill proposal to replan a task or close a task set

Draft task packets, readiness notes and clarifying questions for multi-agent plans

Verify one agent execution round and write a machine-readable result file

Draft task-level detail packets that a multi-agent workflow can import

Brainstorm and converge on decisions with independent takes from several agents

Hand unfinished work to another agent with enough context to take it over

Coordinate notices and shared-file conflicts across parallel agent sessions

Run a multi-expert debate with a structured insight card and formal report

Verify a handoff's claims against sources before taking an irreversible action

Split a large task into roles and parallel sub-threads and merge the results

Write self-contained task descriptions when handing work to a sub-agent or session

Run headless agent sessions over the Agent Client Protocol with multi-agent flows

Create reusable subagent definitions with a role prompt and tool selection

Split multi-part work across parallel subagents when the parts are independent

Delegate features, PR reviews and refactors to Codex, Claude Code or Pi agents

Harden a plan with independent Claude and Codex review, then build and inspect it

Pick a suitable model and a scoped handoff for a task in Claude Code or Codex

Have Codex build a work order while Claude independently inspects the result

Have Codex independently review an implementation plan in a bounded revision loop

Start the Comet workflow and route to Native or Classic mode

Start or resume the Classic Comet workflow for a recoverable change

Start or resume a change in the Comet Native workflow

Offload mechanical coding chores like renames and boilerplate to a cheaper Qwen subagent

Formulate and delegate tasks to local CLI agents such as Codex, Cursor and OpenCode

Run the Loopy agent loop toolkit under its older loop-library name

Find, audit and run repeatable AI agent loops for recurring work

Coordinate reviewer, editor and tester Claude Code sessions with a judge rubric

Look up how the Ralph orchestrator works and debug its failure modes

Create and validate Ralph hat collections for multi-agent workflow routing

Run, monitor and debug autonomous Ralph loops and their merge queues