n1-insight-strategy-refiner

name: n1-insight-strategy-refiner description: LP、記事LP、漫画LP、広告、バナー、ショート広告の上流で、N1の既存認知、信念の穴、脳内競合、形成すべき新認知、本能インサイト、CV直前抵抗を洗い出し、offer改善とRAG投入可能な戦略パックまで作るSkill。 metadata: trigger_keywords: - N1インサイト - 既存認知 - 新認知 - 本能に刺す - 需要のニュアンス - マーケ戦略ブラッシュアップ - n1-insight-strategy-refiner

n1-insight-strategy-refiner

関連資産

• このSkill固有:
  • references: .agents/skills/n1-insight-strategy-refiner/references/ (instinct-insight-rubric.md, n1-cognition-map.md, offer-objection-loop.md, rag-output-schema.md)
  • templates: .agents/skills/n1-insight-strategy-refiner/templates/ (insight_objection_map.json, n1_cognition_strategy.json, rag_chunks_template.json, strategy_pack_template.md)
  • agents: .agents/skills/n1-insight-strategy-refiner/agents/openai.yaml
• 共有:
  • 共通script: tools/run_with_gui_env.zsh, tools/dpro_mcp_bridge.mjs, tools/gemini_model_policy.mjs
  • 共有reference: knowledge/skill-routing.md, knowledge/operating-principles.md, knowledge/production-gates.md, knowledge/workflows/rag_to_creative_pipeline.md
  • 共有template: templates/work_order_minimum.yaml, templates/blocked_handoff.md
  • 運用ドキュメント: docs/agent_operating_doctrine.md, docs/request_templates.md

Use this skill when the owner wants to sharpen marketing strategy before LP, article LP, manga LP, banner, short ad, or ad copy production. This skill is the common upstream N1 cognition and insight OS.

It does not generate Image2 assets, final scripts, HTML, uploads, or publishing artifacts. It produces strategy, offer decisions, and RAG-ready chunks that downstream production skills can use.

Core Contract

• Start from N1 psychology, not demographics. Define what the N1 currently believes, what they want to believe but cannot, and what mental competitor blocks action.
• The minimum N1 schema is active_n1, current_belief, belief_gap, brain_competitor, purchase_or_click_anxiety, desired_new_cognition, new_cognition_trigger, proof_need, final_resistance, and forbidden_claims.
• Always use serious-mode objection testing. For real strategy refinement, the three skeptical N1 critics are separate subagents when the runtime supports subagents. Do not treat single-agent roleplay as the default quality bar.
• Treat external market, DPro, competitor, SNS, Reddit, X, and Meta observations as market_observation only. Do not use them as CVR, sales, profit, No.1, or guaranteed outcome proof.
• Treat agent synthesis as creative_hypothesis unless it is grounded in owner-owned results. Treat owned conversion/sales/customer data as owned_outcome only when explicitly provided.
• Always separate copy changes from offer, product, CTA, funnel, pricing, guarantee, or legal changes. Human-side changes must be listed clearly.
• A good new cognition is not a feature explanation. It changes the N1's internal model of the problem, risk, identity, or next action.
• The final output must be usable by another agent without re-deciding the strategy.

DPro Objective Classification Policy

• DPro正式分類は dpro_ad_objective_raw として保存し、実制作で使う分類は canonical_ad_objective として分ける。
• 分類根拠は dpro_objective_classification に残す。
• 原則はDPro MCPの ad_objective を正とし、transition_type、LP形状、CTA、遷移URLだけで AWARENESS を獲得扱いにしない。
• 違和感がある場合のみ、ad_objective_override と ad_objective_override_reason をセットで残す。理由なしmanual overrideは無効。
• demo_offer / デモ住宅診断 / デモ診断オファー / example-offer.test / example-offer.test 関連itemは例外として、DPro側が AWARENESS でも canonical_ad_objective=acquisition とする。
• N1逆算では、DPro上の獲得/認知/unknownを分け、認知広告を獲得N1の証拠として扱う場合は必ず仮説扱いにする。

When To Use

Use this skill when the request includes any of:

• "N1起点でブラッシュアップ"
• "既存認知 / 新認知を洗い出したい"
• "インサイトを捉えたい"
• "本能に刺したい"
• "今すぐ買わない理由がないまで詰めたい"
• "広告/LP/漫画LP/バナーの上流戦略を作りたい"

Use this before manga-lp-strategy-rebuild when the work needs broad N1 cognition strategy first. Use manga-lp-strategy-rebuild after this when the next step is manga LP story/page strategy.

Required Source Map

Before strategy synthesis, build a source map with evidence states:

• owned_fact: official product, LP, form, pricing, offer, owner-approved assets.
• owned_outcome: actual owned CV, sales, user interview, sales call, customer feedback, ad result.
• market_observation: DPro, competitor LPs, Meta Ads Library, X, Reddit, YouTube, search, public reviews.
• learning_rag: promoted prior learnings only.
• creative_hypothesis: agent-generated N1 or copy hypothesis.
• missing_info: facts needed for stronger strategy.

If evidence is missing, continue only with conservative assumptions and mark the limitation.

Workflow

1. Resolve target and boundaries.
   • Identify project/product, target output, current offer, target CV, price, CTA, and allowed scope of change.
   • If the user allows product changes, still separate them from copy and presentation changes.
2. Build the N1 cognition map.
   • Read references/n1-cognition-map.md when current belief, belief gap, brain competitor, or new cognition is central.
   • Produce at least five desired_new_cognition candidates and score them.
3. Extract instinct-level insights.
   • Read references/instinct-insight-rubric.md.
   • Evaluate at least eight instinct axes, including safety, love/protection, loss avoidance, pride/status, belonging, self-efficacy, guilt avoidance, and future recovery.
4. Run direct-sales objection loops.
   • Read references/offer-objection-loop.md.
   • Spawn three skeptical N1-adjacent subagents when available: busy buyer, anxiety-heavy buyer, and value skeptic.
   • Treat the owner's standing preference for this skill as consent to use this serious-mode subagent loop whenever the task invokes n1-insight-strategy-refiner.
   • If subagents are unavailable in the runtime, record FALLBACK_MAIN_AGENT_ROLEPLAY in objection_direct_sales_rounds.md and apply the same critic prompts with extra skepticism.
   • Iterate until the remaining objections are either resolved, accepted as human-side changes, or recorded as blockers.
5. Decide offer/product/CTA changes.
   • Separate copy_change, offer_presentation_change, product_change, pricing_change, cta_flow_change, proof_change, and legal_or_owner_confirmation.
6. Create strategy artifacts.
   • Use templates/strategy_pack_template.md for the owner-readable pack.
   • Use JSON templates when machine-readable output is needed.
7. Build RAG-ready chunks.
   • Read references/rag-output-schema.md.
   • Emit rag_n1_strategy_chunks.json with source/evidence metadata.
   • If MARAG runtime is available and the task asks for ingestion, ingest into domain=market_strategy and write rag_n1_strategy_ingest_report.json. If unavailable, mark BLOCKED_RUNTIME_UNAVAILABLE but keep chunks.

Standard Artifacts

Produce these in the project output directory unless the owner asks for a lighter answer:

• n1_cognition_strategy.md
• n1_cognition_strategy.json
• insight_instinct_map.md
• objection_direct_sales_rounds.md
• offer_product_change_decision.md
• creative_strategy_brief.md
• rag_n1_strategy_chunks.json
• rag_n1_strategy_ingest_report.json

Quality Gates

Fail or revise the strategy when:

• N1 is only demographic.
• desired_new_cognition is just a feature or benefit restatement.
• instinct insight is generic, polite, or surface-level.
• brain competitor is missing.
• final resistance before CTA is missing.
• copy changes and product/offer changes are mixed.
• market observations are used as outcome proof.
• downstream production would need to decide the core promise, N1, CTA, or forbidden claims again.

Handoff Defaults

• LP/article LP handoff: provide page/block-level reader-state jobs and CTA readiness conditions.
• Manga LP handoff: provide belief shift, character conflict, opening hook, proof beats, and manga-lp-strategy-rebuild next-step prompt.
• Banner/ad handoff: provide hook axes, instinct, visual metaphor, forbidden claims, and CTA intent.
• Short ad handoff: provide first 2-second hook, problem reframe, proof type, and ending action.

Resource Guide

• references/n1-cognition-map.md: use for current belief, belief gap, brain competitor, and new cognition.
• references/instinct-insight-rubric.md: use for deep emotional and instinct-level insight scoring.
• references/offer-objection-loop.md: use for skeptical N1 direct-sales loops and offer/product refinement.
• references/rag-output-schema.md: use when producing or ingesting RAG chunks.
• templates/strategy_pack_template.md: owner-readable output skeleton.
• templates/n1_cognition_strategy.json: machine-readable N1 cognition skeleton.
• templates/insight_objection_map.json: instinct and objection map skeleton.
• templates/rag_chunks_template.json: RAG chunk skeleton.