ericcgladstone-maker

Experiment Simulation Presentation — AI skill for Claude Code

AI community

Building and simulating a behavioral experiment with an AI agent: a real Claude Code session, console on the left, commentary on the right.

How to install Experiment Simulation Presentation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open ericcgladstone-maker/experiment-simulation-presentation and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Experiment Simulation Presentation does

Building and simulating a behavioral experiment with an AI agent: a real Claude Code session, console on the left, commentary on the right. Replies and tool output verbatim; the organization and its brief are fictional.

Alternatives in AI

  • CodeIsland — Real-time AI coding agent status panel in your MacBook notch — live status, approvals & replies for 13 AI tool 2.3k ★
  • Mirofish CLI — Multi-agent AI prediction engine - digital sandbox for scenario simulation (fork of 666ghj/MiroFish) 307 ★
  • Vera Eb Suite — Open-source AI skills and plugins for EB-1 and EB-2 NIW immigration petitions — from evidence organization to 219 ★

README

Building and simulating a behavioral experiment with an AI agent

Eric Gladstone · research walkthrough · independent work · September 2026. The presentation's own header reads "Experimental rehearsal · Experimental design".

A real Claude Code session, presented as it was worked through. The Claude Code console is on the left and my commentary is on the right. Starting from a client brief, the session decides what the first experiment has to answer, reviews outside evidence, specifies the measures and decision thresholds, compares designs, and builds the chosen one as protocol, analysis plan and code. It then simulates the experiment 400 times in each of ten synthetic worlds where the truth is known, stress-tests its assumptions, finds where it could mislead a decision, revises it, retests the revision on the same worlds, and ends with a conditional recommendation.

What is verbatim and what is editorial

The questions are mine. Claude's replies and tool output are verbatim, from one continuous session (Claude Code 2.1.282, Opus 5.5), recorded 27 September 2026. Turns 1–20 are the original build. Turns 21–23 are later questions in the same session, labeled on screen, that show one synthetic experiment end to end, draw two figures from the saved results, and bring the decision table forward; they changed nothing the earlier turns produced. Omitted parts of replies are marked on screen. Timing, grouping into parts, focus, and the commentary on the right are editorial.

The organization (Civic Access Network) and its brief are fictional, written for this demonstration. The session ran in an isolated directory, and account names are removed from the published data.

Use

Open it and step with ← →. Use shift+← → to move between sections and space to play. A− / A+ (or the - and = keys) change the terminal text size. URL options: `