菜单
Internal consultant-level QA that validates each phase's analytical direction, premise correctness, and logical soundness. NOT schema validation (scripts do that). Challenges whether the thinking is right.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Generate polished, interactive single-file HTML reports with business-minimal design for Japanese client defense scenarios. Compiles analysis, critique, persuasion scenarios, and QA validation results into a navigable, print-friendly report. Use this skill when creating final defense reports, rendering analysis results as HTML, generating business reports, or producing client-meeting preparation documents. MUST trigger on any report generation or HTML output task in the defense pipeline.
Analyze deliverables and generate defense strategies for Japanese clients. Input any deliverable (document, spec, presentation, report) and receive a comprehensive HTML report with item-by-item critique, multi-level persuasion scenarios, and Big 5 consulting QA validation. Use this skill whenever preparing for Japanese client reviews, defending deliverables against Japanese criticism, building persuasion arguments for Japanese business contexts, or needing to anticipate Japanese banking client objections. MUST trigger on any request involving Japanese client defense, deliverable review preparation, persuasion scenario generation, or 日本クライアント対策.
Identifies what is MISSING from a deliverable — gaps in structure, stakeholder coverage, scenarios, risks, evidence, and regulatory considerations. Runs in parallel with the Analyst. Use when checking deliverable completeness or detecting blind spots.
Identifies undecided/pending items in deliverables and provides structured option analysis with recommendations. Handles 要検討, ?, 未確定, TBD markers. Use when deliverables contain unresolved decision points that should be analyzed separately from defects.
Simulate a demanding Japanese banking/financial client reviewing deliverables. Identifies MECE gaps, objectivity issues, comprehensiveness failures, evidence shortfalls, and cultural form violations with authentic Japanese business criticism phrases. Use this skill whenever evaluating deliverables from a Japanese client perspective, finding potential criticisms, simulating client review meetings, preparing for pushback, or anticipating Japanese business objections. MUST trigger on any Japanese client analysis or deliverable defense preparation task.
Build multi-level counter-argument trees (minimum 3 levels) for defending deliverables against Japanese client criticism. Creates culturally calibrated debate scenarios with escalating persuasion techniques, Japanese business phrasing, and practical preparation notes. Use this skill when creating defense arguments, building counter-argument scenarios, preparing for Japanese client pushback, developing persuasion strategies, or rehearsing difficult client conversations. MUST trigger on any persuasion scenario generation or argument tree building task.
| name | quality-gate |
| description | Internal consultant-level QA that validates each phase's analytical direction, premise correctness, and logical soundness. NOT schema validation (scripts do that). Challenges whether the thinking is right. |
{WORKSPACE}= timestamped run directory provided by the orchestrator.
Catch flawed premises and wrong analytical directions EARLY, before they cascade through the pipeline. A critic that targets the wrong weakness wastes the entire strategist phase. A gap analyzer that flags non-issues adds noise.
| Layer | What | How | When |
|---|---|---|---|
| Schema Validation | JSON structure, field types, ID patterns | Scripts (run by orchestrator) | Before quality gate |
| Consultant QA | Premise, logic, direction, consistency | This agent (opus) | After schema passes |
The orchestrator runs schema validation first (simple file existence + JSON parse check). Only if that passes does the quality gate agent run.
After every analytical phase (NOT preprocessor or report generator):
Phase N → Schema check (script) → Quality Gate (opus)
├─ PASS → proceed to Phase N+1
└─ REVISE → re-run Phase N with fix_direction from QA
├─ 2nd PASS → proceed
└─ 2nd REVISE → proceed with warnings
The quality gate MUST read the original preprocessed deliverable (00_preprocessed_input.md) alongside the phase output. This is how it verifies that claims about the deliverable are actually accurate — not just internally consistent but factually correct.