map-reduce

name: "map-reduce" description: "When it would yield better results to explore mulitple approaches to a problem, load this skill to orchestrate multiple agents using a diverge and converge strategy."

::types::

TYPE Task = instructions for a sub-agent to execute TYPE Strategy = "accumulate" | "independent"

::input::

$transform: Task $validator: Task $strategy: Strategy = "accumulate" $maxIterations: Number = 5

::context::

$statusQuo: Link $iteration: Number = 0

::workflow::

USE ~/skill/work-packages TO create master work package

DO Summarise status quo into $statusQuo END

IF $strategy = "accumulate" THEN SPAWN ~/agent/general TO run accumulate prompt WITH #accumulate-prompt ELSE SPAWN ~/agent/general TO run independent prompt WITH #independent-prompt END

RETURN $statusQuo

Accumulate Prompt

WHILE $iteration < $maxIterations AND NOT diminishing returns DO
  AWAIT SPAWN ~/agent/general TO $transform WITH
    input: $statusQuo
    output: next candidate path
  AWAIT SPAWN ~/agent/general TO $validator WITH
    candidate: next candidate path
  IF validation passed THEN
    $statusQuo = next candidate path
  END
  $iteration = $iteration + 1
END

RETURN $statusQuo

Independent Prompt

WHILE $iteration < $maxIterations DO
  ASYNC SPAWN ~/agent/general TO $transform WITH
    iteration: $iteration
    output: candidate path
  $iteration = $iteration + 1
END

AWAIT SPAWN ~/agent/general TO $validator WITH
  candidates: candidate paths from async runs

RETURN winning candidate