expert-validator

Expert Validator

Validate your research strategy with 3 independent expert agents. Uses Task Agents for multi-perspective evaluation. Enriches strategy-brief.md with consensus, divergence, and confidence ratings.

Purpose

Expert Validator answers the question "Is this strategy REALLY good, or just looking good?"

A single AI perspective creates blind spots. This skill spins up 3 specialized Task Agents — each with a fresh context window, independent evaluation, and distinct expertise — then synthesizes where they agree and disagree.

Where 3 experts agree → Strong signal. Act on it.
Where experts diverge → Decision point. You choose.
What everyone missed → Blind spot found. Investigate.

The output enriches research-memory/strategy-brief.md with [expert-validator] tagged sections for Expert Consensus, Expert Divergence, and Confidence Overview.

"Consensus = signal in noise." — The Boring Marketer (Expert Review Framework)

Prerequisite

research-synthesizer must have run first. This skill reads strategy-brief.md for the synthesized strategy to validate. If the file doesn't exist or has no content beyond the scaffold, stop and instruct the user to run research-synthesizer first.

Additional research-memory files (market-landscape, competitive-intel, customer-insight, customer-language) are loaded as supporting context for the expert agents.

Modes

Mode When to Use Behavior

Full Validation First run, or strategy-brief.md has no [expert-validator] sections All 3 agents evaluate the entire strategy

Focused Validation User has a specific question or decision point All 3 agents focus on one question

Re-Validation [expert-validator] sections exist but research-memory was updated Re-run agents and update Expert sections

Auto-Load Protocol

On every invocation, BEFORE any evaluation:

Check research-memory/ directory
If files exist → Read ALL .md files (except README.md)
Critical: Read strategy-brief.md
If file missing → STOP. Tell user: "Run research-synthesizer first to generate the strategy brief."
If file exists but has no substantive content → STOP. Same instruction.
Check brand-memory/ (read-only) → If exists, include business description and positioning context in agent briefings
If [expert-validator] sections already exist in strategy-brief.md → suggest Re-Validation mode
Summarize what's loaded and confirm mode with user

Input Gathering

Collect conversationally. Most inputs come from research-memory — just confirm with the user.

Field Required Description

Validation mode YES Full / Focused / Re-Validation

Focus question Focused mode only The specific question or decision to validate

Business context Optional Current stage, resource constraints, timeline — sharpens agent evaluations

Language Optional 결과물 작성 언어 (default: English)

If this is Full Validation, confirm: "I'll have 3 expert agents review your entire strategy brief. Proceed?"

If this is Focused, ask: "What specific question or decision do you want the experts to evaluate?"

If this is Re-Validation, show the current Expert sections and ask: "Research was updated. Should I re-run all 3 experts?"

Process

Step 1: Prepare Agent Briefing Packet

Goal: Compress all research-memory context into a focused briefing that each agent receives.

Build the briefing packet from loaded files:

Briefing Packet

Business Overview

[From brand-memory/ or user input — what the business does, stage, constraints]

Market Context (from market-landscape.md)

Market category: [definition]
Market size: TAM [X], SAM [X]
Key trends: [top 3 with opportunity/threat tags]

Competitive Context (from competitive-intel.md)

Competitive set: [direct + indirect competitors]
Key gaps/opportunities: [from competitive analysis]

Customer Context (from customer-insight.md + customer-language.md)

Primary segment: [description]
Top pain points: [top 3]
Key customer language: [top phrases/expressions]

Strategy Brief (from strategy-brief.md — FULL TEXT)

[Include the complete strategy brief content — this is what agents evaluate]

Validation Focus

[Full: "Evaluate the entire strategy." / Focused: "Specifically evaluate: [user's question]"]

Keep the packet under 3000 words — agents work better with focused context than raw dumps.

Step 2: Run 3 Expert Agents (Sequential)

Launch 3 Task Agents sequentially. Each receives the same briefing packet but evaluates from a distinct perspective.

IMPORTANT: Do NOT pass one agent's output to the next. Each agent must evaluate independently with a fresh context.

Agent 1: Growth Strategist

Task tool call:

Task( subagent_type: "general-purpose", description: "Growth Strategist evaluation", prompt: [see Agent Prompt Template below, filled for Growth Strategist] )

Perspective: Market entry timing, Go-To-Market direction, growth levers

Evaluation questions:

Is the market entry timing right given market maturity and trends?
Is the recommended GTM approach realistic given competitive intensity and resources?
What is the single strongest growth lever available?
What growth opportunity did the strategy miss?
What is the biggest growth risk?

Agent 2: Brand Strategist

Task tool call:

Task( subagent_type: "general-purpose", description: "Brand Strategist evaluation", prompt: [see Agent Prompt Template below, filled for Brand Strategist] )

Perspective: Positioning opportunity, messaging angles, tone direction

Evaluation questions:

Does the identified positioning align with actual market gaps?
Does the messaging reflect how customers actually talk? (cross-check customer-language)
Is the differentiation point clear and defensible vs. competitors?
Is the brand tone appropriate for the target audience?
What positioning opportunity did the strategy miss?

Agent 3: Customer Acquisition Expert

Task tool call:

Task( subagent_type: "general-purpose", description: "Acquisition Expert evaluation", prompt: [see Agent Prompt Template below, filled for Acquisition Expert] )

Perspective: Channel priorities, early traffic strategy, quick wins

Evaluation questions:

Do the recommended channels match where the audience actually spends time?
Is the early traffic/lead strategy realistic for the business stage?
What can be done in 30 days for a quick win?
Is the approach CAC-efficient for this business model?
What channel or tactic did the strategy miss?

Agent Prompt Template

Use this template for all 3 agents. Fill [ROLE] , [PERSPECTIVE] , and [QUESTIONS] per agent.

You are a [ROLE] with 15+ years of experience in [PERSPECTIVE].

Your Briefing

[INSERT FULL BRIEFING PACKET FROM STEP 1]

Your Task

Evaluate the strategy brief from your specialized perspective.

Answer these 5 questions: [INSERT 5 EVALUATION QUESTIONS]

Output Format (follow EXACTLY)

Strengths

[2-3 strategy elements you agree with. Be specific — reference data from the brief.]

Concerns

[2-3 issues or risks. Explain WHY with evidence from the brief.]

Missing

[1-2 blind spots the strategy overlooked. What should have been considered?]

Recommendation

[Your single most important recommendation. One sentence, actionable.]

Confidence

[High / Medium / Low] — [One sentence explaining your confidence level]

Rules

Be SPECIFIC and ACTIONABLE — no generic advice
Reference actual data from the briefing (market numbers, competitor names, customer language)
If you disagree with a recommendation, explain WHY with evidence
Do NOT hedge everything — take clear positions
Write your evaluation in [user's specified language]. If no language specified, use English.
Keep total output under 400 words

Step 3: Synthesize Consensus & Divergence

Goal: Analyze all 3 agent outputs and extract signal from noise.

After collecting all 3 evaluations, synthesize directly (no additional agents needed):

3a. Expert Consensus

Scan all 3 outputs for agreement:

Strong Signal ⭐ (3/3 agree): All three experts highlight the same strength, concern, or recommendation
Moderate Signal (2/3 agree): Two experts align on a point

For each consensus point:

State the point clearly
Note which agents agree
Summarize the shared reasoning

3b. Expert Divergence

Scan for conflicting positions:

Identify points where agents disagree or contradict each other
Present BOTH sides with their reasoning
Note the implication for decision-making — what does the user need to decide?

3c. Confidence Overview

Compile confidence ratings:

Expert Confidence Key Reason

Growth Strategist [H/M/L] [one-line reason]

Brand Strategist [H/M/L] [one-line reason]

Acquisition Expert [H/M/L] [one-line reason]

Overall [H/M/L] [synthesized judgment]

Overall confidence rule:

3× High = High
2× High + 1× Medium = High
Mixed = Medium
Any Low = Medium (flag the concern)
2+ Low = Low (strategy needs rework)

3d. Recommendations Reinforcement

Review the existing Strategic Recommendations in strategy-brief.md :

Which ones are validated by expert consensus? Mark them.
Which ones are challenged? Note the concern.
Are there NEW recommendations from the experts? Add them.
Re-prioritize based on consensus strength.

Step 4: Save & Log

Goal: Write expert sections to strategy-brief.md and log execution.

4a. Enrich strategy-brief.md

Language rule: 섹션 헤더와 테이블 컬럼명은 영어로 유지합니다. 본문, 셀 값, 설명, 분석 텍스트는 사용자가 지정한 언어로 작성합니다. 언어가 지정되지 않으면 English로 작성합니다.

Add or update these sections with [expert-validator] tags. Do NOT delete any existing content — only add/update expert sections.

Expert Consensus

Source: [expert-validator] | Validated: [YYYY-MM-DD]

Strong Signals (3/3 agree)

[consensus point] — Growth ✓ Brand ✓ Acquisition ✓

[shared reasoning summary]

Moderate Signals (2/3 agree)

[consensus point] — [Agent A] ✓ [Agent B] ✓

[shared reasoning summary]

Expert Divergence

Source: [expert-validator] | Validated: [YYYY-MM-DD]

[Topic of disagreement]

[Agent A]: [position + reasoning]
[Agent B]: [opposing position + reasoning]
Decision needed: [what the user should decide]

Confidence Overview

Source: [expert-validator] | Validated: [YYYY-MM-DD]

Expert	Confidence	Key Reason
Growth Strategist	[H/M/L]	[reason]
Brand Strategist	[H/M/L]	[reason]
Acquisition Expert	[H/M/L]	[reason]
Overall	[H/M/L]	[synthesized]

Also update ## Strategic Recommendations — append [expert-validator] annotations to existing items and add new expert-sourced recommendations.

For Re-Validation: Replace existing [expert-validator] sections entirely. Append > Re-validated: [date] to each section header.

4b. Update research-log.md

Append one row:

Quality Checklist

Before saving, verify:

All 3 agents ran independently (no cross-contamination of outputs)
Consensus section identifies at least 1 Strong Signal or 2+ Moderate Signals
Divergence section is populated (if all 3 agree on everything, note "No significant divergence")
Each consensus/divergence point references specific data (not generic)
Confidence overview includes all 3 agents + overall rating
Existing strategy-brief.md content is preserved (enrichment only)
All expert sections have [expert-validator] source tags
Strategic Recommendations updated with expert annotations
research-log.md updated with execution record

Example (Abbreviated)

Input: Full Validation of marketing education business strategy.

Agent 1 (Growth Strategist):

Strengths: AI marketing education timing is strong — creator economy CAGR 12-15%
Concerns: $199 price point faces heavy competition from free content; conversion path unclear
Missing: No paid acquisition strategy as growth lever
Confidence: Medium — timing good, but monetization path needs work

Agent 2 (Brand Strategist):

Strengths: "Boring" positioning is sharp differentiation in hype-heavy AI market
Concerns: "Boring" may conflict with premium perception if expanding upmarket
Missing: No voice-of-customer language validation on the "boring" resonance
Confidence: High — positioning angle is clear and defensible

Agent 3 (Acquisition Expert):

Strengths: SEO + newsletter first strategy fits solo-operator resources
Concerns: YouTube absence is the biggest missed channel — top search engine for tutorials
Missing: No referral or affiliate strategy for low-CAC growth
Confidence: Medium — channel mix is too narrow

Strong Signal ⭐: "AI-fatigued practitioner" is the right primary segment (3/3) Strong Signal ⭐: SEO is the right anchor channel (3/3) Divergence: Pricing — Growth says lower entry + upsell, Brand says hold premium position Overall Confidence: Medium — strategy direction solid, execution gaps need addressing

What This Skill Does NOT Do

Generate strategy → Use research-synthesizer (creates the strategy brief this skill validates)
Conduct new research → Use market-scanner , competitor-finder , audience-profiler , etc.
Execute marketing → Use execution skills (brand-voice, copy, email, SEO, etc.)
Replace human judgment → Expert agents provide perspectives; the user makes the final call

Expert Validator is a quality gate — it stress-tests strategy before execution begins.