Japanese-Client-Breaker

English | 简体中文 | 繁體中文 | 日本語 | 한국어

Japanese-Client-Breaker is a defense harness for high-stakes Japanese enterprise reviews, where approval risk, indirect objections, and internal justification matter.

This project was used in real work, including work to persuade Japanese banks and insurance companies. It was built for situations where delivering the document is only half the job, and the harder half is surviving a high-stakes review, answering indirect objections, and giving stakeholders a face-saving rationale they can defend internally.

Status: under active construction.

What This Is

Japanese-Client-Breaker is a practical defense harness built for one specific job:

turn a deliverable into a defense strategy that can stand up to a high-stakes Japanese enterprise review.

It is designed for review cultures where the real pressure is not only content quality, but also:

• MECE discipline
• comprehensiveness
• evidence quality
• precedent sensitivity
• formal consistency
• risk awareness
• face-saving communication

This is not a generic prompt pack. It is a structured pipeline for analyzing a deliverable, predicting criticism, building rebuttal trees, quality-checking the arguments, and packaging the result into a usable report.

Why It Exists

In high-stakes Japanese enterprise reviews, a good deliverable still fails if:

• the logic has even a small gap
• the categories are not MECE
• the rationale is not explicit enough
• the reviewer cannot justify the recommendation internally
• the reviewer feels approval creates personal risk

This harness exists to prepare for that reality.

Quick Start

This repository is designed to be used from inside Claude Code. You do not need to wire the agents manually.

1. Clone the repository and move into it.

git clone https://github.com/HHC225/Japanese-Client-Breaker.git
cd Japanese-Client-Breaker

2. Put your deliverables in input/.

Supported inputs include Excel, PDF, PowerPoint, Word, CSV, images, and plain text files. If one deliverable spans multiple files, put all of them in input/.

3. Launch Claude Code from the repository root.

If your local installation uses the default CLI entrypoint, this is typically:

claude

4. Paste this prompt.

Analyze the deliverables in `input/` for a high-stakes Japanese enterprise review. Run the full defense pipeline and generate the HTML report.

5. Open the generated report at _workspace/latest/defense-report.html (or _workspace/run_YYYYMMDD_HHMMSS/).

Requirements

• Python 3.10+
• uv for the preprocessing script. It can auto-install on first run, but if that fails, install it manually:

curl -LsSf https://astral.sh/uv/install.sh | sh

• Internet connection for first-run dependency resolution and Google Fonts in the HTML report. The report falls back to system fonts if offline.

How It Works

The current design has five core agents:

1. deliverable-analyst Breaks the deliverable into critique-ready items with assumptions, dependencies, and structural gaps.
2. jp-client-critic Simulates a high-stakes Japanese enterprise reviewer and generates realistic objections.
3. persuasion-strategist Builds multi-level counter-argument trees with culturally calibrated phrasing.
4. consulting-qa Stress-tests the arguments through McKinsey, BCG, Bain, Deloitte, and Accenture lenses.
5. report-generator Compiles the full output into a practical HTML report for meeting preparation.

Repository Structure

• input/ Put the deliverable files you want to analyze here.
• _workspace/ Generated at runtime. Each run creates a timestamped subdirectory (run_YYYYMMDD_HHMMSS/) with intermediate artifacts and the final HTML report. A latest symlink points to the most recent run.
• CLAUDE.md Runtime instructions for using this repository inside Claude Code.
• .claude/agents/ Agent definitions for the end-to-end defense pipeline.
• .claude/skills/ Reusable skills, references, and orchestration logic.
• japanese-client-characteristics-research.md Source research on Japanese banking and financial client behavior.

Best Fit

This project is best suited for:

• teams preparing for high-stakes Japanese enterprise reviews
• delivery teams working under approval pressure inside Japanese organizations
• strategy and PM teams that need recommendations people can justify internally
• operators building agent workflows for objection handling and approval support

License

This repository is released under the MIT License.

Use it freely.

Contributing

If this harness helps you survive a high-stakes Japanese enterprise review, answer indirect objections, or build approval-ready arguments people can defend internally, add your pattern, edge case, phrasing, or improvement and open a pull request.

Good field experience is more valuable than polished theory. Please contribute it back.

Contributions are especially welcome if you can add:

• real objection patterns from Japanese client reviews
• better face-saving phrasing
• stronger persuasion trees
• sharper QA standards
• more realistic financial-sector edge cases

The goal is simple: make this more useful in real meetings.