writing-an-orchestrator-fan-out-promptlisted
Install: claude install-skill sblattj/agent-chandlery
## More triggers
swarm-prompt (old name), write a prompt for parallel sub-agents, generate a research swarm prompt.
# writing-an-orchestrator-fan-out-prompt: author a fan-out orchestrator prompt
Use when the user wants a prompt (text they hand to another agent) that spawns a swarm of
parallel sub-agents and then merges the results. The output is the prompt, not the swarm run,
unless they ask you to launch it. Default behavior: build the prompt, scrub internal
identifiers, save it to a file, print it.
> In plain words: this makes the instructions for a "manager" agent that splits a job across
> several "worker" agents at once and staples their answers together. You get the manager's script.
## Procedure
1. **Name the job in one sentence.** "research issue-tracker workflow best practices", "review
this diff for bugs across dimensions", "compare 4 caching libraries". If you cannot state
it, ask (see Clarifying questions).
2. **Pick the shape** from the Shape library below (research / review / analysis / build).
The shape decides lane structure, per-agent contract, and synthesis format.
3. **Gather grounding context** so lanes are tailored, not generic. Then **scrub** it (see
Scrub rules). Scrub is on by default; only keep an identifier if the user says to.
4. **Decompose into lanes** that are MECE (mutually exclusive, collectively exhaustive) and
non-overlapping. Keep the count sane: match it to real sub-topics and to cost, not to a
round number. Note the