Composing System Prompts begins the series, followed by Curating Knowledge Collections. Open either workshop from the landing page. Light and dark modes are available across the workshop pages. Configuring Skills and Tools completes the series with a 13-slide workshop.
- Composing System Prompts · Full copy
- Curating knowledge collections · Full copy
- Configuring skills and tools · Slides PDF · Participant exercises
- Presenter lesson plans and access requirements
- Latest copy refinements and before/after
- Source history
Workshop 1 requires individual access and Sandbox sign-in. Workshop 2 adds Workspace and Knowledge collection access. Workshop 3 adds Skills and Tools access, with Workspace authoring permissions for creation and editing. Knowledge access is needed in Workshop 3 when a chosen procedure retrieves from a collection.
Participants learn how user prompts and system prompts differ, then compare how two models interpret a sentence about a nurse and doctor. They ask whether to walk or drive to a car wash. After reading sample system prompt instructions, participants locate System Prompt in Chat Controls, add those instructions, and regenerate responses to their original prompt. Participants compare responses, check citations, and test tools in teaching and research tasks. After reviewing Workspace model cards, participants choose STEM Adventure Games for teaching or Compare Wikipedia Edits for research, clone it, and revise its instructions. Workshop 2 can also be taken independently. Participants clone either example, choose their own source documents, create a collection, revise system prompt references, and test responses against source passages. Workshop 3 continues participants’ saved custom models. Skills guide repeatable procedures; tools execute operations. Participants add and test each, then revise either resource or create their own with CAIL Workspace Support. Adventure setting, goals, and rules come from each model’s system prompt.
- STEM Adventure tool and playable preview
- Extend STEM Adventures skill
- System prompt
- Prism Laboratory scenario, aperture variation, and winning commands
- Additional knowledge entries
- Skill Creator instructions and Tool Creator instructions
Both creators accept requirements for any suitable task. STEM Adventure is a workshop request, not a default in either creator.
The engine checks moves, inventory, and prerequisites. Exported records can be restored by replaying commands. Type discuss to place a run in the Sandbox message box for review and sending. A game’s programmed observations are distinct from historical evidence.
Static HTML, CSS, and JavaScript. Published pages have no runtime dependencies. Install the pinned Markdown package to regenerate and check reading pages.
python3 -m pip install -r requirements-build.txt
python3 -m http.server 8766
python3 scripts/check_series.py
python3 scripts/test_copy_regressions.py
node scripts/test_deck_interactions.cjs
node examples/adventure/test-engine.cjs
python3 examples/adventure/test-validation.pyAfter changing game sources, run python3 examples/adventure/build_tool.py to rebuild the Open WebUI tool and preview. After changing slide text, run python3 scripts/check_series.py --write to update the complete transcript, per-session mirrors, and direct copy diffs. Review the generated diff before committing. The checker protects retained source passages and verifies local links, screenshot hashes, accessible slide labels, and article-free mini-agendas.
Use arrow keys, the slider, or Outline to navigate. On mobile, the slider occupies a full row above the navigation buttons. Text selection does not advance slides. Screenshot slides reserve the viewport for the image, heading, and caption. Clicking an image expands it. Outline links to both available sessions, full workshop copy, and Model Registry. Presenter lesson plans remain available through this README.
Developed by Zach Muhlbauer from system-prompting, knowledge-collections, and skills-tools. Original repositories remain available.
Platform instructions follow the CUNY AI Lab Sandbox documentation. Current UI screenshots were captured in Firefox. Reused comparison screenshots exclude obsolete controls. Capture provenance and source limitations remain in review files.