Building Tabletop Exercises for Incident Response Using Codex as Game Master

---

Background and challenge

Tabletop exercises (TTX) for incident response are widely adopted by SRE and operations teams to measure and improve response capabilities through simulated scenarios. The Coincheck Tech Blog case demonstrates a system where Codex, an AI coding agent, takes on the game master (GM) role. This reduces reliance on human facilitators while making exercises easier to run regularly.

📑Table of Contents

1. Background and challenge
2. How Codex acts as game master
3. Preparation flow and sample timeline
4. Measured effect and cost
5. Adoption steps and cautions
6. FAQ
7. Human GM vs Codex GM
8. Summary

The background challenge was that traditional tabletop exercises required experienced facilitators. Preparing scenarios, ideal timelines, question sets, and answer collections manually consumed significant time and made frequent execution difficult. Coincheck addressed this by leveraging Codex’s reasoning capabilities to automate much of the GM workload.

---

How Codex acts as game master

The solution centers on the open-source repository at https://github.com/Kuniwak/ttx. Codex acts as the game master with reasoning effort set low for fast responses, managing the exercise flow via Slack or a TUI application. Preparation involves generating incident causes, ideal timelines, and expected Q&A sets from FTA/FMEA analyses or existing runbooks using AI skills. Generated materials undergo automated review to prevent leakage of sensitive information.

---

Preparation flow and sample timeline

A sample timeline starts at 2026-06-08 10:03:00 with a checkout error rate spike alert. Observability through Prometheus/Grafana shows rising errors in checkout and payment services. The exercise then simulates triage to determine whether the root cause lies in payment processing, network issues, or dependencies. Multiple such timelines are prepared for repeated practice.

---

Measured effect and cost

Measured effectiveness includes strong internal feedback describing the exercises as “highly realistic” and “stress-inducing in a good way.” Development took 11 business days using Claude Code, with per-exercise costs of 100-200 yen and a fixed monthly cost of around 8,000 yen on GCE. SREs and web engineers can adopt the approach with relatively low effort.

---

Adoption steps and cautions

Implementation begins by cloning the repository, configuring the Codex API key, and preparing scenario templates along with review scripts. Key precautions include thorough pre-checks for sensitive data in generated content and iterative prompt tuning to maintain response quality. Start with a small team, gather feedback, and refine scenarios accordingly.

---

FAQ

Common questions include: the main benefit of using Codex is reduced preparation effort and higher repeatability; required preparation materials are incident cause lists, ideal timelines, and Q&A sets; costs run approximately 100-200 yen per session with low fixed monthly fees; differences from human GMs are 24/7 availability and consistent responses; precautions center on mandatory security reviews of generated scenarios.

---

Human GM vs Codex GM

Item	Human GM	Codex GM
Cost	High	Low (100-200 yen/session)
Availability	Low	High
Reproducibility	Medium	High
Development period	–	11 business days

Source: Coincheck Tech Blog (https://tech.coincheck.blog/entry/codex-ttx) (as of June 2026)

---

Related articles:

• Dynatrace Expands AI Coding Agent Monitoring with OpenTelemetry for Claude Code, Gemini CLI, Codex CLI
• Codex app 26.609 Update
• OpenAI Acquires Ona to Strengthen Codex Cloud Environments

Summary

In summary, using Codex as game master for tabletop exercises significantly lowers preparation overhead while enabling high-quality, repeatable training. SRE and operations teams considering adoption should begin with small-scale pilots and customize the system to their environment. Refer to the original Coincheck Tech Blog article and the GitHub repository for full details.

Related new article:

• ChatGPT Work Explained: Long-Running Agents and the Codex Desktop Merge – This published update adds current operational context for Building Tabletop Exercises for Incident Response Using Codex as Game Master.