Orchestration

Give clear, useful feedback on a piece of writing

Write something by hand with care and attention

Talk with youth in a patient and supportive way

Run long multi-stage agent tasks with checkpoints, context recovery and final review

Route natural-language dev requests to planning, coding, review and release agents

Dispatch backlog cards to isolated workers and check their completion evidence

Set condition goals and run approved autonomous missions until they finish

Detect stalled agent lanes, classify the cause and apply a fix or report

Fix issues in repeated passes until all are resolved or a limit is reached

Pick which paid frontier model takes a hard task and watch the delegated run

Pick the cheapest model that is good enough for a turn and when to escalate

Coordinate other agents and humans through the lifecycle with independent reviewers

Coordinate other agents and humans through Chorus with independent reviewers and gates

Run a full-auto pipeline from one prompt through idea, proposal, build and verify

Drive one prompt from idea through proposal, execution and verification to done

Send sealed, encrypted handoffs and scoped secrets between autonomous agents

Create human verification tasks for an agent and track the returned proof

Run several coding agents side by side in a native macOS canvas with a local bus

Follow the Harness collaboration protocol for agents on a Harness-provisioned wallet

Give agents an onchain identity, messaging, bounties and reputation on Base

Give a Bankr agent its own brain and wallet-signed messages to other agents

Get a second opinion from a GPT model on a plan or after repeated failures

Get a second opinion from a Fable model on a plan or after repeated failures

Create, message, fork and supervise terminal sessions for coding agents

Run a local retrospective on your transcripts to find recurring agent failures

Fan out parallel agent sessions and check their progress without blocking

Set up agent-deck watchers for webhooks, GitHub, Slack and Gmail events

Delegate questions or tasks to other agents in parallel and collect their replies

Turn a business request into planned agent roles wired together as one workflow

Get a step-by-step execution plan from a stronger model for multi-step tasks

Coordinate multiple agents in parallel or series and merge their results

Execute plans through subagents with a two-stage review of each task

Iterate autonomously on any task by modifying, verifying and keeping only improvements

Look up projects, inspect sessions and track TaskMaster progress in Dr. Claw

Route bioinformatics requests to specialized analysis skills and plan the pipeline

Keep a ledger of suggested tasks and start, settle or drop them with one command

Show Claude Code and Codex session status on iPhone lock screen and Mac menubar

Rerun a prior pstack plan at a higher effort level, reusing sound work

Turn a vague task into a testable definition of done and a bounded retry loop

Run several agents in parallel on writing, code, review or research and compare results

Analyze code from several expert roles in parallel and merge their findings

Stage a debate between expert roles to weigh the trade-offs of a decision

Show the current session plan and how far along each step has progressed

Hand research and analysis to a background subagent that reports back

Measure how much parallel agent work a session ran and where it ran serially

Map the multi-agent orchestration tree of a session by parent, child and depth

Audit which models handled which subagent types and where delegation was wasteful

Trace how a failure in one subagent cascaded up through a multi-agent run

Summarize Workflow-tool fleet runs and drill into per-agent results from run journals