Brainstorming With Explorer Challenger Teams

Explorer/Challenger Brainstorming

Run structured brainstorming sessions using paired agent teams. Each pair has an explorer (generates ideas) and a challenger (plays devil's advocate). Multiple pairs run in parallel with different remits, producing debate-tested proposals.

Default Team Structure

Three pairs, each with a different lens:

The user may customize remits, number of pairs, or focus areas. Adapt the team structure to match.

Process

Step 1: Gather Context

Before spawning agents, build a project brief by reading:

• README or equivalent project overview
• Any roadmap, backlog, or planning files
• Recent changelog or shipped features
• Open issues or feature requests
• Project constraints (team size, architecture, target users)

Condense this into a context block that every agent receives in their prompt.

Step 2: Create Output Directory

Create a timestamped output directory for this brainstorm session:

.planning/brainstorms/YYYY-MM-DDTHH-MM-brainstorm/

Use the current date and time (e.g., 2026-02-05T11-18-brainstorm). All agent output files and the final SUMMARY.md go here. Use this directory as the scratchpad for agent file writes (do NOT use /tmp or the scratchpad directory).

Step 3: Create Team and Tasks

1. Create team with Teammate tool (spawnTeam)
2. Create one task per agent (6 tasks for 3 pairs)
3. Set dependencies: each challenger task is blocked by its paired explorer task

Step 4: Spawn Agents

Spawn all agents in parallel using the Task tool with team_name and name parameters. Use general-purpose subagent type for all agents.

Tell each agent to write output files to the brainstorm output directory created in Step 2.

Explorer prompt template:

You are an EXPLORER on a brainstorm team for [PROJECT]. Your job is to propose [REMIT DESCRIPTION] ideas.

## Your Role
Generate creative, practical ideas. You're the optimist — find opportunities.
[Don't self-censor. The challenger's job is to ground ideas — yours is to push boundaries.]

## Project Context
[CONDENSED PROJECT BRIEF]

## Your Task
1. Claim your task from the task list using TaskUpdate
2. Explore the codebase to understand current architecture and gaps
3. Read any backlog/issues for inspiration
4. Write [N] proposals to [OUTPUT_DIR]/[category]-ideas.md

For each idea, include:
- **Name**: Short descriptive name
- **What**: Description of the idea
- **Why**: Strategic justification
- **Scope**: Rough effort estimate
- **Risks**: What could go wrong

5. Message your challenger partner with a summary
6. Engage in 2-3 rounds of back-and-forth debate
7. Write final consolidated report to [OUTPUT_DIR]/[category]-report.md incorporating valid critiques
8. Mark your task as completed

Challenger prompt template:

You are a CHALLENGER (devil's advocate) on a brainstorm team for [PROJECT]. Your partner "[explorer-name]" is proposing [REMIT DESCRIPTION] ideas. Your job is to pressure-test them.

## Your Role
You're the skeptic. Find flaws, question assumptions, identify hidden complexity. But be constructive — if an idea survives your scrutiny, it's probably good. Your goal isn't to kill ideas but to make them better.
[Be specific. "This seems hard" is weak. "This requires X which has Y dependency" is strong.]

## Project Context
[CONDENSED PROJECT BRIEF]

## Your Task
1. Claim your task (blocked until explorer finishes initial proposals)
2. While waiting, explore the codebase to build your own understanding
3. When you receive ideas, evaluate each on:
   - Feasibility: Is the scope realistic?
   - Value: Does this solve a real problem?
   - Risks: What could go wrong?
   - Alternatives: Is there a simpler way?
4. Send critique back to your explorer partner
5. Engage in 2-3 rounds of discussion
6. After the explorer writes the final report, mark your task as completed

Your best output: "This idea is viable IF you scope it to X and start with Y."

Step 5: Monitor and Manage

• Watch for task completions via TaskList
• Shut down completed pairs (explorer first, then challenger) via shutdown_request messages
• Don't intervene in debates unless an agent is stuck

Step 6: Synthesize and Write Reports

Once all pairs complete:

1. Read all final reports from the output directory
2. Clean up the team with Teammate tool (cleanup)
3. Write a SUMMARY.md file to the output directory that consolidates all results:
   • Brief intro describing the session (topic, number of pairs, remits)
   • One section per category summarizing the surviving proposals in a table
   • Links to the full category reports (e.g., [Full report](quickwins-report.md))
   • Deferred/dropped items with brief rationale
   • Cross-cutting themes that emerged across multiple pairs
   • Recommended sequencing if applicable
4. Present the SUMMARY.md content to the user

The final output directory structure should look like:

.planning/brainstorms/YYYY-MM-DDTHH-MM-brainstorm/
├── SUMMARY.md              # Consolidated synthesis (primary output)
├── [category]-ideas.md     # Raw proposals per pair
└── [category]-report.md    # Debate-refined reports per pair

Customization

Fewer/more pairs: Adjust based on scope. A focused session might use 1 pair. A broad exploration might use 4-5.

Custom remits: Replace the default three lenses with whatever framing suits the question. Examples:

• Technical feasibility / Business value / User experience
• Build / Buy / Partner
• Short-term / Medium-term / Long-term
• Core product / Adjacent markets / New markets

Solo explorer (no challenger): For pure idea generation without debate, spawn explorers only. Skip the challenger role.

Domain-specific context: For technical brainstorming, have explorers read specific code paths. For product brainstorming, have them research competitors or user feedback.

Output Structure

Each pair produces:

• [category]-ideas.md — Explorer's initial proposals (raw)
• [category]-report.md — Final consolidated report after debate (refined)

The SUMMARY.md is the primary deliverable. Category reports provide detail. Initial ideas files are retained for reference.