B
Offline
Step 1 of 10

Define Your Roster

3of 10~10 min
  • Role-based agent orchestration
  • The three canonical Brenner roles
  • How to configure role-to-agent mapping
What you'll do
  • Understand the three Brenner roles
  • Map each agent to a role
  • Configure the roster in brenner CLI

The Brenner method uses role specialization to ensure productive disagreement. Each agent gets a role-specific prompt that emphasizes different operators and disciplines.

Why roles matter: Without role specialization, all three agents tend to converge on similar ideas. Roles force productive tension — the hypothesis generator proposes, the test designer challenges falsifiability, and the critic attacks the framing.

The Three Canonical Roles

Hypothesis Generator

Codex (GPT)

Primary Operators

⊘ Level-split⊕ Cross-domain◊ Paradox-hunt

Produces

Hypothesis slate with 2-5 candidates + third alternative

Key Discipline

No mechanism = no hypothesis

Test Designer

Claude (Opus)

Primary Operators

✂ Exclusion-test⟂ Object-transpose🎭 Potency-check

Produces

Discriminative tests that separate hypotheses

Key Discipline

Every test needs a potency check

Adversarial Critic

Gemini

Primary Operators

ΔE Exception-quarantine† Theory-kill⊞ Scale-check

Produces

Attacks on framing + assumption challenges

Key Discipline

Attack the framing, not just the details

1Configure the Roster

The brenner CLI accepts a role map that assigns each agent identity to a role. Agent identities in Agent Mail use adjective+noun names (like BlueLake, PurpleMountain).

Role Mapping
# Example role map for brenner session start
--role-map "BlueLake=hypothesis_generator,PurpleMountain=test_designer,GreenValley=adversarial_critic"
# Breaking it down:
# BlueLake     → Codex (GPT)   → hypothesis_generator
# PurpleMountain → Claude (Opus) → test_designer
# GreenValley  → Gemini       → adversarial_critic

2Agent Identity Names

Agent Mail uses adjective+noun names for agents. You'll pick names when registering each agent. Choose memorable names that help you track which agent is which.

Codex Agent

Suggested: BlueLake, GreenMeadow

Claude Agent

Suggested: PurpleMountain, VioletForest

Gemini Agent

Suggested: GreenValley, OrangeDesert

Alternative Role Mappings

The default mapping works well for most research questions. But you can experiment:

DomainHypothesis GenTest DesignCritic
Biology/MedicineCodexClaudeGemini
Computer ScienceClaudeCodexGemini
Social ScienceGeminiClaudeCodex
PhilosophyClaudeGeminiCodex
Pro Tip
Document your role mapping in a session notes file. This helps you track which configuration worked best for different research domains, and makes sessions reproducible.

Roster Ready? Pick your agent names and role assignments. In the next step, you'll write the kickoff prompt that seeds all three agents.