Release: Table Top Agent Skill

Something small to help preparedness...

Hey everyone,

I've been running tabletop exercises with clients and my teams for years. Ransomware scenarios. Threat actor simulations. AI-specific attack chains. And honestly? Most (not all) tabletop consultations are underwhelming. You spend a bunch of money on a “reputable” consulting firm to come up with lackluster insights, expectations, and an event that leaves much to be desired.

So we built something to help.

Introducing: The Arcanum Tabletop Exercise Generator

Our TabletopExercise skill generates professional, interactive HTML handouts for tabletop exercises. No more copy-pasting from Word docs. No more vague scenarios. Just solid, executable exercises that actually test your team's readiness.

What It Generates

  • Professional HTML handouts — self-contained files for ease of use for participants

  • Facilitator version — Full script, timing, expected responses, facilitation notes

  • Participant version — Scenario, timeline, injects, discussion questions — no spoilers

  • Realistic artifacts — Fake but believable logs, alerts, emails, and dashboards to show during the event. Generated by AI but passable to satisfy the technical people in the room.

  • Interactive gap analysis — Fillable evaluation forms with owner/due date tracking

  • Detailed Injects — Timed scenarios with evidence, discussion questions, and facilitator guidance

An Example Scenario: Ransomware on Developer Network

Prompt to the Skill: “Create a tabletop for an incident where ransomware has infected a portion of the developer network…”

Scenario Built: Ransomware encrypts GitLab repositories, Jenkins build servers, and developer workstations. The attackers demand $500,000 in Bitcoin with a 72-hour deadline. They've also exfiltrated the source code.

This scenario tests:

  • Incident response process and playbooks

  • Network segmentation (dev vs. prod)

  • Backup recovery procedures

  • Ransomware payment decisions

  • Executive communication

  • Regulatory notification

Output: HTML Facilitator guide, HTML participant Guide, Five injects, 5 Inject artifacts, Gap analysis guide.

The difference between a plan that works and a plan that looks good in a PDF is whether you've tested it. With the right people.

Now you can generate a professional tabletop exercise in minutes. Not for compliance. For actual preparation.

How to Use It

Load the skill and describe your scenario:

  • Load skill folder into your agent frameworks skill folder: TabletopExercise. We use claude code; so .claude/skills

  • Prompt: "Ransomware attack on development network"

    • (more detailed descriptions help the skill better output)

The skill generates both HTML files. Share the participant version with attendees. Keep the facilitator version for yourself. You can describe a scenario or a threat actor. If the 1st generation isn't to your liking, just prompt Claude or your model to make the adjustments, then update the skill file to reflect your needs for next time.

What's Included

  • Skill file and a script to take AI research from JSON to HTML guides.

(Sponsor)

Cut pentest reporting time. Automate findings delivery in real time.

PlexTrac helps teams graduate from spreadsheet tracking to a repeatable findings-to-fix workflow: standardize intake, reduce duplication, keep evidence attached, and push actions to the right owners. Less chasing. More closure.

Get a demo HERE

(Jason Note: We love Plextrac, check them out!)

Where to Get It

Free on the Arcanum GitHub here. The skill is available now.

Another Scenario (and the template included in the skill)

Prompt: “Create a tabletop for a DDoS attack on the game Rainbow Six”

Some example screenshots:

Outro + Disclaimer

This skill worked wonderfully on Claude Code Opus 4.5. We then tested it using the MiniMax m2.5 model. Model does really matter sometimes. Our m2.5 versions were a little less great, but still totally workable with some additional prompting. Most of the issues were with HTML formatting or the level of detail in the AI's data gathering for the scenario. Many portions of this skill were vibe-coded. Some were hand-created as well.

Consider this skill IN EXTREME BETA and a TEMPLATE for you to edit and make your own.

We will be out at the B-Sides Seattle and NDC Oslo conferences over the next few weeks. If you are around, come say hi!

— Jason