3-Round Brainstorm: Claude x Gemini (v2.1 — Two-Layer Architecture)

Setup (first run only)

Before first brainstorm, verify these prerequisites:

1. Find

   gemini.py

   — bundled at

   scripts/gemini.py

   relative to this SKILL.md. Locate it with:

   find . ~/.claude/skills -name gemini.py -path '*/gemini/*' 2>/dev/null | head -1
   

   Store the result as

   GEMINI

   variable for the session.

2. Check

   GOOGLE_API_KEY

   — must be set. Check with

   echo $GOOGLE_API_KEY

   . If empty, look for

   .env

   file and source it, or tell the user to get a key at https://aistudio.google.com

3. Check

   google-genai

   package — run

   python3 -c "from google import genai; print('OK')"

   . If missing, run

   pip install google-genai

   .

If any check fails, tell the user what's missing and how to fix it. Do NOT proceed without all three.

Two-layer model architecture: Flash researches, Pro reasons. Claude orchestrates both.

                  Claude (orchestrator)
                     |         |
        Phase 0.5    |         |    R1 / R2 / R3
        Phase 3.5    |         |
                     v         v
              Flash-Lite --grounded    Pro (no grounding)
              (research layer)    (reasoning layer)
                     |                  ^
                     +--Verified Context-+

Flash-Lite (

gemini-3.1-flash-lite-preview

gemini-3.1-pro-preview

Version History

• v1: 3 rounds, no web access. Result: 2/6 decisions invalidated by stale knowledge (A488).
• v2: Added

  --grounded

  to all Pro calls. Problem: Pro wastes thinking tokens on search processing.
• v2.1: Two-layer split. Flash does research, Pro focuses on reasoning. Faster, cheaper, better quality.

Why two layers beat one

In v2, Pro with

--grounded

Splitting the layers means:

• Flash-Lite (grounded): $0.25/$1.50 per 1M in/out, 3-5 sec per query, returns verified facts
• Pro (ungrounded): all thinking tokens go to critical analysis, gets pre-verified context

Process

Phase 0: Gather Context

Before writing Round 1, check if the user provided enough context. A good R1 prompt needs:

• What exists (product, tech, assets)
• Hard constraints (time, team size, budget, regulatory)
• The goal (what decision needs to be made)

If any of these are missing, ask the user before starting. A brainstorm with vague input produces vague output.

Phase 0.5: GROUNDING (dual research — Claude + Flash)

This phase is mandatory. Extract all technology names, frameworks, APIs, and services from the context.

Two parallel research tracks:

1. Claude runs 5-7 WebSearch queries (versions, pricing, compatibility, licenses)
2. Flash (grounded) gets a batch research prompt covering the same technologies

Run both in parallel. Merge results into a "Verified Context" block.

Flash research prompt (saved to

/tmp/brainstorm-{agent}-ground.txt

Research the following technologies and provide CURRENT (2026) facts for each.
For each one, include: current version, pricing/free tier, license, known issues.
Use Google Search to verify — do NOT rely on training data.

Technologies to research:
1. [Technology A]
2. [Technology B]
3. [Technology C]
...

Output format per technology:
- Name: current version, release date
- Pricing: free tier details, paid plans
- License: open-source? which license?
- Compatibility: works with [other stack items]?
- Red flags: any known issues, deprecations, EOL dates

Flash call:

python3 $GEMINI ask @/tmp/brainstorm-${AGENT}-ground.txt \
  -m gemini-3.1-flash-lite-preview --grounded --save /tmp/brainstorm-${AGENT}-ground-response.md

Merge: Claude reads Flash's response + own WebSearch results, resolves conflicts (if any), produces final Verified Context block:

## Verified Context (dual-checked: Claude WebSearch + Flash Google Search)
- Next.js: v16.1.6 (Oct 2025 release, 16.2 canary). Source: nextjs.org, confirmed by Flash
- Clerk: Free 10K MAU. Pro $25/mo. @clerk/nextjs@6.36.7. Source: clerk.com
- Drizzle: v0.45.x, zero-codegen, Neon HTTP driver. Source: drizzle.team
- CONFLICT: [if Claude and Flash disagree, note both and flag for R1]

Phase 1: DIVERGE (Round 1)

Claude writes a prompt with:

1. Verified Context from Phase 0.5 (at the top)
2. Full context (what exists, constraints, goals)
3. Initial honest assessment (strengths, weaknesses)
4. Instructions for Gemini to challenge, not agree

Pro is called WITHOUT

--grounded

R1 prompt template:

You are acting as an adversarial brainstorming partner. I am the lead developer framing the problem.
This is Round 1 of a 3-round brainstorm.
Your job is NOT to be agreeable -- challenge, flip assumptions, add angles I haven't considered.

IMPORTANT: The "Verified Context" section below has been web-checked moments ago.
Treat these facts as ground truth. Focus your energy on STRATEGY and ARCHITECTURE,
not on verifying versions or prices — that's already done.
If you want to recommend a technology NOT in the verified list, flag it clearly
so we can verify it before Round 2.

## Verified Context (web-checked)
[Include Phase 0.5 merged findings]

## Context
[Describe the idea, what you have, constraints, goals]

## My Initial Assessment
[Your honest evaluation: strengths, weaknesses, open questions]

## What I want from you (Round 1)
1. Challenge my framing -- what am I NOT seeing?
2. Propose 3-5 alternative angles/applications
3. Kill at least 1 idea I proposed (with reasoning)
4. Add 1 wildcard idea I haven't considered
5. If you recommend a NEW technology not in Verified Context, flag it for verification

After R1: If Gemini recommended new technologies not in Verified Context, run a quick Flash grounded check on those before proceeding to R2. Tell the user what new angles Gemini generated.

Mid-round verification (only if R1 introduced new technologies):

python3 $GEMINI ask \
  "Verify these technologies: [new tech from R1]. Current version, pricing, license, compatibility with [our stack]." \
  -m gemini-3.1-flash-lite-preview --grounded --save /tmp/brainstorm-${AGENT}-r1-verify.md

Phase 2: DEEPEN (Round 2)

Claude reads R1 response, then writes R2 prompt that:

1. Summarizes Gemini's ideas (including any mid-round verification results)
2. Kills ideas with concrete arguments (synthesize, do NOT copy-paste raw R1 output)
3. Adds real-world constraints (time, solo dev, regulatory, etc.)

Pro is called WITHOUT

--grounded

R2 prompt template:

This is Round 2 of our brainstorm. Here's what you said in Round 1, followed by my critical evaluation.
Your job now: STRESS-TEST my pushback, pick the 2 strongest surviving threads, and kill everything else.

All technology claims have been web-verified. Focus on architecture and strategy.

## Your Round 1 Key Ideas (summary)
[Summarize Gemini's R1 ideas — synthesize, don't copy-paste]

## My Critical Evaluation
[For each idea: KILL or KEEP with concrete reasoning]

## Constraints you must respect
[Time, resources, regulatory, skill gaps]

## What I want from you (Round 2)
1. Where am I wrong in my kills? You get ONE sentence per killed idea to defend it.
2. Pick the 2 strongest surviving threads (from ideas I kept).
3. For each survivor: concrete 2-week plan.
4. Propose 1 new idea that combines surviving threads.

After R2 — USER CHECK-IN (mandatory): Present the 2-3 surviving ideas to the user and ask: "Do you have a preference before we lock one in, or should I proceed to Round 3?" This prevents the models from converging on something the user doesn't care about.

Phase 3: CONVERGE (Round 3)

Claude synthesizes both rounds, raises one remaining objection. Incorporate user preference from check-in if given.

Pro is called WITHOUT

--grounded

R3 prompt template:

This is Round 3 (FINAL) of our brainstorm. Time to converge.

## What survived 2 rounds
[List 2-3 artifacts with descriptions — synthesize, don't copy-paste]

## My objection
[Raise one remaining concern]

## What I want from you (Round 3 -- FINAL)
1. Address my objection -- defend or kill.
2. Pick ONE primary action (not 2, not 3 -- ONE).
3. Sequence: primary -> secondary -> kill list.
4. Write the 1-sentence pitch for the winner.
5. Give a concrete timeline (hours/days, not weeks).
6. Define go/no-go signal.

Phase 3.5: FACT-CHECK (dual verification of final decisions)

After R3, extract all technology decisions, pricing claims, and compatibility assumptions from the final recommendation.

Two parallel verification tracks (same as Phase 0.5):

1. Claude runs WebSearch on each claim
2. Flash (grounded) gets a verification prompt with all claims

Flash fact-check prompt:

Verify each of these claims using Google Search. For each one, state:
CONFIRMED (with source) or INCORRECT (with correction and source).

Claims to verify:
1. [Claim from R3]
2. [Claim from R3]
...

Output format (merged from both tracks):

If any claim is invalidated or conflicted, flag it and note the correction. Include the fact-check table in the final synthesis.

Phase 4: Synthesize

Read all 3 responses + fact-check results and produce:

1. Summary table:

2. Final recommendation — the ONE action with timeline
3. Fact-check results — merged verification table
4. Kill list — everything considered and rejected (with reasons)

Execution

File naming convention

All temp files MUST include the agent ID to prevent collisions when multiple agents brainstorm in parallel.

Pattern:

/tmp/brainstorm-{agent}-{phase}.txt

/tmp/brainstorm-{agent}-{phase}-response.md

Where

{agent}

rnd

cmo

seo-geo

qa

Model assignment

gemini-3.1-flash-lite-preview

gemini-3.1-pro-preview

gemini-3.1-flash-lite-preview

gemini-3.1-flash-lite-preview

Sending prompts

Setup: Before first call, locate

gemini.py

1. .claude/skills/brainstorm/scripts/gemini.py

   (project-level)

2. ~/.claude/skills/brainstorm/scripts/gemini.py

   (global)

3. ~/.claude/skills/gemini/gemini.py

   (separate gemini skill)

GOOGLE_API_KEY

.env

export

Agent ID: Use project name or a short identifier to prevent file collisions. Store in variable

AGENT

# Example — adapt paths to your install location
GEMINI="path/to/gemini.py"  # resolved at session start
AGENT="myproject"

# Phase 0.5: Flash-Lite grounded research
python3 $GEMINI ask @/tmp/brainstorm-${AGENT}-ground.txt \
  -m gemini-3.1-flash-lite-preview --grounded \
  --save /tmp/brainstorm-${AGENT}-ground-response.md

# R1: Pro reasoning (ungrounded)
python3 $GEMINI second-opinion @/tmp/brainstorm-${AGENT}-r1.txt \
  --save /tmp/brainstorm-${AGENT}-r1-response.md

# R1.5: Flash-Lite verify new tech (only if R1 introduced unverified tech)
python3 $GEMINI ask \
  "Verify these technologies: [new tech from R1]. Current version, pricing, license, compatibility with [our stack]." \
  -m gemini-3.1-flash-lite-preview --grounded --save /tmp/brainstorm-${AGENT}-r1-verify.md

# R2: Pro reasoning (ungrounded)
python3 $GEMINI second-opinion @/tmp/brainstorm-${AGENT}-r2.txt \
  --save /tmp/brainstorm-${AGENT}-r2-response.md

# R3: Pro reasoning (ungrounded)
python3 $GEMINI second-opinion @/tmp/brainstorm-${AGENT}-r3.txt \
  --save /tmp/brainstorm-${AGENT}-r3-response.md

# Phase 3.5: Flash-Lite fact-check
python3 $GEMINI ask @/tmp/brainstorm-${AGENT}-factcheck.txt \
  -m gemini-3.1-flash-lite-preview --grounded \
  --save /tmp/brainstorm-${AGENT}-factcheck-response.md

Requirements:

GOOGLE_API_KEY

pip install google-genai

If a Gemini call fails (rate limit, API failure, timeout), retry once. If it fails again, stop and report the error to the user.

Saving artifacts

• All prompts + responses in

  /tmp/brainstorm-{agent}-*.txt

  and

  /tmp/brainstorm-{agent}-*-response.md

• If brainstorm is about a project experiment, copy all artifacts to the experiment directory so they survive /tmp cleanup

Key Principles

1. Flash researches, Pro reasons — never waste Pro's thinking tokens on factual lookups.
2. Web-ground before reasoning — Phase 0.5 is mandatory. Stale facts poison all 3 rounds.
3. Dual verification — Claude WebSearch + Flash Google Search in parallel. Conflicts get flagged.
4. Claude challenges between rounds — not relay. Write kills with arguments, not just summaries.
5. Synthesize, don't copy-paste — R2 and R3 prompts should distill previous rounds, not dump raw output.
6. Mid-round verify — if Pro suggests a new technology in R1, Flash checks it before R2.
7. Kill early, kill often — R1: 7+ ideas, R2: cut to 3, R3: converge to 1.
8. Constraints are ammunition — real limits (time, solo dev, budget) kill impractical ideas.
9. One action, not a strategy — R3 must end with a single deliverable + timeline.
10. Fact-check closes the loop — Phase 3.5 catches anything grounding missed.
11. Go/no-go criteria — every brainstorm ends with "if X doesn't happen by Y, abandon."
12. Show the user the journey — present the funnel (7 -> 3 -> 1) so the user sees why the winner won.
13. Additional rounds when needed — 3 rounds is the default, not the limit. After R3 convergence, add R4+ rounds when: (a) the Go/No-Go test produces results that need Gemini evaluation, (b) new information invalidates R3 assumptions, or (c) the fact-check reveals a critical correction that changes the recommendation. Additional rounds follow the same Pro (ungrounded) model. Label them R4, R5, etc. with clear purpose: "R4: test results validation", "R5: pivot evaluation".