Workshop Notebooks
This area is restricted to workshop attendees. The instructor announced an
access key during the session โ enter it below, or type
roadshow --key <key> in the terminal.
{ notebooks }
Each notebook is self-contained. Open the rendered HTML to read in
place, or download the .ipynb and run it yourself with
the same commands the workshop uses.
The full repo (notebooks, the .claude/ spec-driven
workflow, sample data) is open source. Follow along on a real
terminal:
$ git clone https://github.com/joshuamschultz/ai-roadshow.git
$ cd ai-roadshow
$ uv sync
$ cp .env.example .env # add ANTHROPIC_API_KEY (or your provider)
$ uv run jupyter lab
No uv? pip install -e the four arc
packages from ../arc/packages/ โ instructions in the
repo README.
chat to coworker
System prompts, message threads, model swap. The model is interchangeable; the harness is what matters.
prompts that work
Good vs bad prompts, the five anti-patterns, and the plan ยท do ยท check pattern that makes agents self-correcting.
the loop
One Python tool, arcrun.run(), streaming events. The loop is just LLM + tools + memory.
skills and prompts
A SKILL.md folder with references. The agent loads only what it needs. Reusable knowledge.
identity-aligned decisions
Conservative vs Aggressive Buyer, same data, different decisions. Tribal knowledge โ identity edit โ diff under git.
full agent audit
Federal-grade tamper-evident event chain. Which version of which identity made which decision.
security at runtime
PII redaction, sandboxed code execution, tool allowlists, PII-safe logging, token/cost budgets. OWASP LLM/ASI top-10 mapped.
coding workflow
The principled-coder identity, then the seven-phase chain end-to-end. Plan/Do/Validate is fractal.
If a viewer link 404s, the rendered HTML hasn't been generated yet โ
run scripts/build-roadshow-notebooks.sh from the repo
root before deploying. The raw .ipynb downloads always
work.