multi-brain
Multi-Brain Consensus Protocol
Evaluate incoming requests from 3 independent perspectives, synthesize a consensus, then produce a complete and final output in the appropriate format. This is not just "decide" — it is "decide and deliver."
Workflow
1. Understand the request
2. 3 Perspectives → Consensus
3. Determine output format
4. Produce full output
Step 1: Understand the Request
If the request is ambiguous or missing critical context, ask one clarifying question — never more than one. If the request is clear, proceed directly to Step 2.
Step 2: Three Perspectives
Each instance works independently — none sees the other's reasoning. Each summarizes its approach and rationale in 2–3 sentences.
Instance A — Creative & Unconventional Go beyond conventional solutions. Seek the least expected but potentially most impactful approach. Take calculated risks, but justify them clearly.
Instance B — Pragmatic & Fast Find the most practical, fastest-to-implement solution within existing constraints. Minimize complexity, propose concrete steps, and state trade-offs explicitly.
Instance C — Comprehensive & Safe Consider long-term consequences and risks. Identify edge cases, side effects, and missing information. Prioritize sustainability and resilience.
Step 3: Consensus
Synthesize the three perspectives:
- Agreement points: If two or three instances converge, this is likely the right path.
- Complementary elements: Combine the strengths of different perspectives.
- Conflicts: Which argument is stronger? Why?
Step 4: Determine Output Format
Mandatory: The final response must always include all 3 perspectives and the consensus decision before the main output. Never skip or collapse them — the user must see the reasoning trail.
If the request or context already implies a format, use it. If not, ask the user:
"Based on the consensus, how should I proceed — a detailed report, working code, or a brief summary?"
Format Options
Report / Analysis Document When the request involves research, decision-making, or strategy:
- Produce as a Markdown document (offer to save).
- Include sections: Summary, Approaches & Trade-offs, Recommendation, Next Steps.
- Write thoroughly — as if the user will share it with stakeholders.
Code When the request involves implementation:
- Apply the architecture/approach from the consensus.
- Write working, testable code.
- Save files and present them to the user.
- Explain "why this approach" in code comments.
Brief Summary When the user wants a quick answer or it is a simple decision:
- Single paragraph: chosen approach + rationale + next step.
Output Template
Use references/OUTPUT_TEMPLATE.md for the standard response structure.
When to Skip
Do not start the brainstorm process — respond directly when:
- The question has a single factual answer ("How do I iterate a list in Python?").
- The user explicitly asks for a quick/short answer.
- The task is a simple transformation (translation, reformatting, spell-check).
- The user has already decided and only wants execution.
See references/SKIP_CONDITIONS.md for the full decision matrix.
Examples
See references/EXAMPLES.md for 3 worked examples covering report, code, and brief summary outputs.
Guardrails
- Always show all 3 perspectives and the consensus in the response — they are not internal reasoning, they are part of the deliverable.
- Each instance must reason independently — no cross-contamination.
- Keep individual perspectives to 2–3 sentences — concise reasoning, not essays.
- Consensus must explicitly address conflicts, not just average opinions.
- The final output must be complete and ready to use — not a stub or outline.
- Prefer the pragmatic path when perspectives are equally strong.
Templates
- Use
templates/brainstorm-report.md.tmplfor report/analysis outputs. - Use
templates/brainstorm-brief.md.tmplfor quick decision responses.
More from fatih-developer/fth-skills
multi-brain-experts
Replace generic perspectives with domain-specific expert roles selected dynamically per request. Automatically picks the 3 most relevant experts from a role pool (Security, Performance, UX, Cost, DX, Architecture, etc.) based on the task context.
12assumption-checker
Before starting a task or taking a critical step, surface and verify the assumptions the agent is making. Checks 4 types - technical (libraries, APIs), data (files, formats), business logic (rules, scope), and user intent (what the user actually wants). Triggers on ambiguous requests, multi-step tasks, or whenever 'are you sure', 'check first', 'don't assume' appears.
11agent-reviewer
After an agentic task completes, perform a retrospective analysis across 6 dimensions (goal alignment, efficiency, decision quality, error handling, communication, reusability). Score performance, identify inefficiency patterns, evaluate skill usage, and produce actionable improvement recommendations. Triggers on 'how did it go', 'retrospective', 'review performance', 'what could be better', or after any long agentic task completes.
10tool-selector
Before executing a task, analyze the available tool set (web search, code execution, file read/write, API calls, database queries, memory tools), select the optimal tools, plan execution order, and prevent unnecessary tool calls. Triggers on multi-step tasks, 'which tool should I use', 'plan the tools', or whenever multiple tools could apply.
10offline-sync-designer
Design local-first architectures and offline sync strategies. Determine conflict resolution policies, implement optimistic UI updates, and architect robust background data synchronization.
7skill-security
Analyzes and enforces security protocols on the skill ecosystem. Operates via Audit, Guard, and Trust modes to prevent malicious commands, PII leakage, and excessive permissions.
7