debate — Challenge, Strategy, Critic

Caution

Opinion without evidence is noise. Every debate starts with a mandatory Discovery phase: parallel research agents scan your codebase and the web. Arguments that don’t cite Discovery findings don’t count — the judge challenges them on the spot.

Tip

Three modes, one command. Challenge picks the best option from competing variants. Strategy synthesizes independent proposals into a plan. Critic finds every weakness in what you already have. The skill auto-asks which mode if you don’t specify.

Quick reference

Field	Value
Command	`/brewtools:debate`
Arguments	`[topic] [-m challenge\|strategy\|critic] [-n 2-5] [-r max-rounds] [--review]`
Model	opus
Agents	2–5 dynamic debaters + secretary + judge (main session)
Output	`discovery.md`, `decisions.md`, `summary.md`, `debate-log.jsonl`

When to use

Scenario	Command
Pick the best technology from several options	`/brewtools:debate "React vs Vue vs Svelte" -m challenge`
Design a migration or architecture plan	`/brewtools:debate "monolith to microservices plan" -m strategy`
Find all risks in an existing solution	`/brewtools:debate "critique our JWT auth flow" -m critic`
Extended analysis with more agents and rounds	`/brewtools:debate "PostgreSQL vs DynamoDB" -n 4 -r 8`
Post-debate code review of output	`/brewtools:debate "our deployment pipeline" -m critic --review`

Modes

Mode	Goal	Agent setup
Challenge (default)	Select the best variant from competing options	Defenders argue FOR, critics attack; judge declares winner
Strategy	Synthesize the strongest approach from independent proposals	All agents are strategists; goal is convergence, not a winner
Critic	Find every weakness, risk, and flaw	All agents are critics with different perspectives; no defender

Examples

# Challenge mode — compare options (default)
/brewtools:debate "React vs Vue vs Svelte for our new dashboard"

# Strategy mode — how to approach a problem
/brewtools:debate "Migration plan from monolith to microservices" -m strategy -n 4

# Critic mode — stress-test an existing plan
/brewtools:debate "Review our authentication flow" -m critic -n 5 --review

Flow

Parse and interview
Extracts topic, mode (-m), agent count (-n), and max rounds (-r) from your command. If mode is omitted, asks explicitly — no auto-guessing. Confirms setup via AskUserQuestion before doing anything else.
Generate agent team
Assigns roles (defender / critic / strategist) based on mode, then selects character archetypes (Pragmatist, Skeptic, Visionary, Architect, Operator, and others) to create productive tension. Displays the full agent table and asks for confirmation before proceeding.
Discovery — mandatory evidence phase
Two research agents run in parallel: Codebase Explorer scans your project for relevant code, patterns, and dependencies; Web Researcher fetches current best practices, official docs, and benchmarks. All findings — with file paths and URLs — are saved to discovery.md and injected into every debater’s context.
Debate rounds
Agents run sequentially; each reads the shared JSONL log before responding. Every argument must reference a Discovery finding — unsourced claims are challenged by the judge. The judge (main session) monitors for consensus, stalemate, or max rounds and can redirect mid-debate.
Summary and decision
A secretary agent writes summary.md with key arguments, turning points, and agreement areas. The judge then writes decisions.md: winning position (or synthesized result), decisive arguments, minority opinions, confidence level, and recommended next steps.
Final output
Status table: mode, rounds used, outcome (consensus / partial / none), top 3–5 decision bullets, and links to all four artifact files. If —review was set, triggers /brewcode:review on the output.

Internals: phases, Discovery rules, JSONL schema, agent prompt composition

Phases (SKILL.md)

Phase	What happens
0 — Validation	Runs `validate.sh`; stops on failure. Loads `archetypes.md` into context.
1 — Parse	Reads `$ARGUMENTS`; blocks until user picks mode if `-m` omitted.
2 — Init	Runs `init-log.sh`; captures `REPORT_DIR` and `LOG_FILE`.
3 — Interview	`AskUserQuestion` to confirm or adjust mode, agent count, rounds, custom profiles.
4 — Agent profiles	Generates team from `setup-flow.md`; user confirms before proceeding.
5 — Discovery	Parallel research agents per `discovery-flow.md`; saves `discovery.md`.
6 — Debate	Mode-specific flow from `challenge-flow.md` / `strategy-flow.md` / `critic-flow.md`.
7 — Summary	Secretary agent writes `summary.md` from full log + discovery.
8 — Decision	Judge (main session) writes `decisions.md`.
9 — Output	Displays status table; optionally invokes `/brewcode:review`.

Discovery evidence mandate

Every argument in Phase 6 MUST cite at least one finding from discovery.md. Format: [Source: #N] where N is the discovery entry number. The judge tracks sourcing — unsourced claims are challenged and do not count toward consensus.

JSONL log schema

Each debate turn is one line in debate-log.jsonl:

{"ts":"2026-04-05T14:30:00","from":"agent-1","to":["agent-2"],"what":"React has larger ecosystem","why":"More packages, hiring pool [Source: #3]","type":"argument","mode":"challenge"}
{"ts":"2026-04-05T14:30:15","from":"agent-2","to":["agent-1"],"what":"Ecosystem size != quality","why":"Vue curated ecosystem avoids dependency hell [Source: #7]","type":"counter","mode":"challenge"}

Entry types: argument, counter, proposal, agree, question, redirect.

Agent prompt composition (Phase 6)

Each agent’s prompt is built dynamically from four layers:

Base template — agents/debater-template.md
Role overlay — agents/{role}-template.md (defender / critic / strategist)
Character traits — from archetypes.md for the assigned archetype
Evidence base — full contents of discovery.md injected as context

Auto-detect keywords (when -m is omitted and topic is unambiguous)

Keywords in topic	Detected mode
compare, choose, vs, best, pick, which, select	`challenge`
strategy, approach, plan, how to, design	`strategy`
critique, weakness, risk, flaw, review, audit	`critic`
(none match)	asks user — never silently defaults

🔍

Updating plugins

Use /brewtools:plugin-update to check and update the brewcode plugin suite in one command. See the FAQ for details.