JHamidun CEO Council — Independent Strategic Analysis
Launch parallel sub-agents as isolated C-level experts. Each analyzes the same project data from their perspective. No coordination between experts — isolation produces genuine diversity of opinion. Then synthesize consensus and disagreements.
Critical: How to Launch Experts
MUST use the Task tool with subagent_type: "general-purpose" and model: "opus".
Task(
subagent_type: "general-purpose",
model: "opus",
prompt: "<expert prompt with data>",
description: "CFO analysis"
)
DO NOT use bash, shell scripts, or background commands to launch experts. They will fail.
Launch all experts in a single message with multiple Task tool calls for true parallelism.
Step 1: Scan Project Context
Before suggesting experts, understand the project:
- Read
CLAUDE.md(orREADME.mdif absent) - Scan
.claude/rules/for domain context - Glance at top-level file structure
Based on findings, generate 4-6 expert roles tailored to THIS project. Roles must reflect the project's actual domain, challenges, and stage.
Step 2: Assemble the Council
MANDATORY: Ask the user before proceeding. Do not pick roles yourself.
Use AskUserQuestion with multiSelect: true:
- Show 4-6 role options with short descriptions of their focus
- User can always pick "Other" to define custom roles
- Minimum 2 experts. If user picks 1, suggest adding one more for productive disagreement
Role Examples by Project Type
Don't copy these — generate fresh roles based on actual project context:
| Project Type | Typical Roles |
|---|---|
| SaaS | Head of Engineering, Head of Product, Head of Growth, CFO, UX Researcher |
| Open Source | Community Manager, Technical Architect, DevRel, Security Advisor |
| Content / Media | Content Strategist, Audience Analyst, Monetization Expert, Distribution Expert |
| EdTech | CMO, CFO, CPO, COO, Growth Advisor |
| E-commerce | Head of Supply Chain, Marketing Director, CTO, Customer Experience Lead |
| Agency / Consulting | Sales Director, Delivery Lead, Talent Manager, CFO |
Step 3: Gather Current Data
Collect project state to feed all experts. Stay focused on what's relevant:
Read:
- Key metrics/data files identified during context scan
- Strategy and planning documents
- Recent decisions or changes (git log --oneline -10)
- Previous council analyses (if any)
Skip: GitHub traffic stats, stargazer counts, clone data, contributor lists — these are vanity metrics, not strategic data.
All experts must receive identical data context. Prepare the data block ONCE, then paste it into each expert prompt.
Step 4: Generate Expert Prompts
For each selected expert, create a prompt with the SAME data block:
You are the [ROLE] for [PROJECT NAME]. Analyze the data below from a [DOMAIN] perspective.
Focus on:
- [3-6 specific focus areas relevant to role and project]
Data:
[CURRENT PROJECT DATA — identical for all experts]
[Role-specific instruction: "show the math", "be the contrarian", "prioritize by effort/impact", etc.]
Respond in the same language as the data provided.
Rules:
- Each expert gets the SAME data block — prepare it once, reuse
- Focus areas must be specific to the project, not generic
- Include a personality instruction (contrarian, pragmatic, data-driven)
- Mention project constraints the expert should know
Step 5: Execute
Launch ALL selected experts in one message using multiple Task tool calls:
# In a single response, call Task for each expert:
Task(subagent_type: "general-purpose", model: "opus", prompt: "<CFO prompt>", description: "CFO analysis")
Task(subagent_type: "general-purpose", model: "opus", prompt: "<CPO prompt>", description: "CPO analysis")
Task(subagent_type: "general-purpose", model: "opus", prompt: "<CTO prompt>", description: "CTO analysis")
Wait for all experts to return results before proceeding to synthesis.
Step 6: Synthesize
Do not skip this step. The synthesis is the entire value of the council.
After all experts report, create a synthesis document:
# Council Session: [DATE]
## Council Members
[List of selected experts and their focus]
## Context
[Current metrics/state snapshot — brief]
## [Expert 1 Name]
[Key findings and recommendations]
## [Expert 2 Name]
[Key findings and recommendations]
## Consensus (all agree)
1. ...
2. ...
## Disagreements
| Expert | Position | Argument |
|--------|----------|----------|
| ... | ... | ... |
## Decisions
_To be filled after discussion._
Consensus Rules (evidence > votes)
- Triage, not averaging. Sort expert outputs into: Consensus (2+ agree) / Expert-only (unique points per expert) / Disagreements. Adjudicate each disagreement explicitly — never split the difference.
- Evidence beats vote count. One expert citing a verifiable number/source from the data block beats three experts agreeing from intuition. All experts run on the same base model — unanimous agreement without evidence is a shared-prior echo (consensus hallucination), not a truth signal. Tag such items
[CONSENSUS-only — no supporting data]in the synthesis. - MODEL_ECHO (if experts run on different models via
gpt-agent/gemini-agent): require first lineMODEL_ECHO=<model id>in each expert reply — catches silent fallback where "diverse council" is secretly one model. Mismatched echo → mark that expert degraded.
Save Results
Save to a logical location:
docs/council-[DATE].md— default- Or project-specific path if context suggests one
Common Mistakes
| Mistake | Fix |
|---|---|
| Picking roles without asking user | ALWAYS use AskUserQuestion first |
| Using bash to launch experts | ONLY use Task tool with subagent_type: "general-purpose" |
| Giving experts different data | Prepare ONE data block, paste into all prompts |
| Gathering vanity metrics | Focus on project docs, strategy, actual metrics |
| Too many experts (6+) | 3-4 is optimal for signal-to-noise |
| Skipping synthesis | The synthesis IS the value — never skip |