59 specialists + 7 workflow-specific orchestrators + 1 generic base
orchestrator. All prompts live in src/agents/prompts/. Each
agent also has a profile entry in profiles.ts (skill ratings,
personality, DiceBear avatar).
Orchestrators (8)
One per workflow + a base. They decide who runs, in what order, and when to stop.
orchestrator.ts — base / generic
orchestrator-counsel.ts — single specialist Q&A
orchestrator-review.ts — specialist + evaluator with revision loop
A .ts file exporting a Prompt string in src/agents/prompts/.
A definition in src/agents/definitions.ts with model tier, max turns, and allowed MCP tools.
A profile in src/agents/profiles.ts with skill ratings + DiceBear seed.
All 67 ride the same frontier LLM (Claude or Mistral). The differentiation
is the system prompt + tool permissions + their slot in the debate
protocol, not separate fine-tuned models.