Start here

Overview Shipped

Every team is a self-contained bundle you install into the registry. Open one to see what it is for, what it can do today, and every specialist and lens inside it.

Teams differ far more than "a group of agents" suggests. Some fan out to a dozen specialists and reconcile the results; some dispatch a single authoring agent against a fixed contract; one runs on a deliberation graph and does not use the objective machinery at all. What they share is the front door, the registry, and the run record.

At a glance

TeamEngineFor
devteamspecialist-pipelineCode review, conformance, planning, and the build loop
teambuilderauthoringCreating new teams and improving existing ones
teamtesterbuiltins onlyRunning the other teams as subjects, at scale
researchteamspecialist-pipelineCited, confidence-scored research on a topic
cookbookteamauthoringRecipes, principles and guidelines from source or intent
storyteamspecialist-pipelineGrounded long-form narrative from a gathered corpus
socialteamauthoringTiered news digests against a fixed interest profile
projectteamroadmapDecomposing and sequencing a feature
brainstormteamdeliberation graphFour-phase facilitated thinking on any idea

Lenses per team

  • brainstormteam146 specialists
  • cookbookteam113 specialists
  • devteam18723 specialists
  • projectteam61 specialist
  • researchteam53 specialists
  • socialteam135 specialists
  • storyteam135 specialists
  • teambuilder31 specialist
  • teamtester00 specialists

devteam dwarfs the rest because code review is the domain that decomposes furthest — there really are a hundred and eighty-seven separable things to look for in a diff. A team with five lenses is not a lesser team; it is a narrower domain.

A note on "specialists"

In a fan-out team a specialist is a real dispatched worker. In an authoring team the specialists are the lenses a single authoring agent has to satisfy — the files say so explicitly, and it would be misleading to count them the same way. teamtester has none at all by design: every one of its commands binds straight to a generic engine builtin, so the judgment lives in the engine's probes and judges rather than in prompts the team ships.

These lists are generated

Every roster below — 47 specialists and 252 lenses, with each one's focus and its verification criteria — is a snapshot built from the team data by a script, not typed by hand. That is the only way a page describing this much surface stays true after the next change to it.