Tools and Skills · Thursday, October 15, 2026 · 2:30–4:00 p.m. Eastern
GC New Media Lab · Room 7388.01
Open Your Model
2:30–2:40 · Slides 1–5
Continue your saved model or clone it first. Open Workspace → Models → Edit and read your actual system prompt. Identify its setting, goals, and rules. Use instructions.txt and source passages when your prompt requires them. Keep this model card open while you add both a skill and a tool. To clone the earlier example, select Workspace in left sidebar, then Models. Search for STEM Adventure Games. Open ⋯ beside it and select Clone.
Add Your Skill
2:40–2:50 · Slides 6–7
Under Skills → Select Skill, choose existing Adventure Panel Play. Read its saved instructions. This orchestration skill coordinates Adventure Panel according to your model's system prompt. Select this existing saved skill.
Add Adventure Panel
2:50–3:00 · Slides 8–10
Stay in the model card during the tool explanation and examples. A tool call sends a function name and arguments for execution; the returned result and rendered output show what actually happened.
Under Filters → Select Filter, enable Model Bound Tool Enabler. It restores attached tools when a chat omits them.
Under Tools → Select Tool, choose existing Adventure Panel. Confirm Adventure Panel Play remains selected under Skills. Check that your model card shows a knowledge collection, tool, skill, and filter. Keep the card open. Add the system-prompt clause below, then save with Save & Update before opening a fresh chat.
Add Skill Instructions
Add this clause to the bottom of your model's system prompt:
Load the attached Adventure Panel Play skill (extend-adventure-games) and follow it on every turn. Use show_scene from the Adventure Panel tool to display each scene and its choices. Follow this system prompt’s setting, goals, rules, and source instructions.
Confirm both Adventure Panel Play and Adventure Panel are selected, then click Save & Update. Your prompt, selected skill, and selected tool must be saved before testing.
Test Your Adventure
3:00–3:15 · Slides 11–12
Run your updated adventure game. Verify that the configured skill and tool are called appropriately, then play the rest of the game. The example run begins with:
initiate adventure
Check the opening menu or scene against your system prompt and any required source passages. Check skill loading, tool execution, and rendered play separately:
| Check | What to inspect |
|---|---|
| Skill | Saved Adventure Panel Play instructions actually loaded |
| Orchestration | Skill coordinates Adventure Panel using your prompt's setting, goals, and rules |
| Tool call | Actual function and supplied arguments match the reviewed adventure |
| Result | Tool returns a usable result and an interactive player appears inside chat |
| Play | A selected choice reaches chat, the next scene follows the supplied rules, and a stop request ends play |
Native discovery may show view_skill; confirm its actual saved-skill name and ID rather than guessing an ID. A direct $ mention is a separate path that injects selected skill content and need not produce a view_skill call. Label which path you used. Merely naming a skill does not establish that its saved instructions loaded.
Open tool-call details and inspect the actual function, supplied arguments, and returned result. Confirm the input follows your model’s prompt and the rendered player appears. Use the schema of your currently attached tool.
Try a choice before and after its stated prerequisite and check the scene against your rules. Select a panel choice. If Open WebUI shows Confirm Prompt from Embed, confirm the numbered choice you intended to send. Check that it appears as your message and that the next panel reflects it. If submission is interrupted, copy the displayed decision into chat and send it once. Ask to stop before the goal is met. Adventure Panel displays each turn and passes the selected choice to chat; its skill applies the supplied rules to established events. These instructions do not guarantee independently enforced game state.
The user-provided live screenshot shows view_skill in a collapsed activity summary, a show_scene call, and a rendered ADVENTURES panel. Open details in your own run to verify the selected skill ID and instructions. The screenshot alone does not show those arguments or prove a completed playthrough.
Explore Other Skills
3:15–3:25 · Slide 13
After your first adventure works, experiment with Adventure Outcomes or Conclude Adventure Game, if your account can select them. Read complete saved instructions and check compatibility with your current tool. Adventure Outcomes instructions are available for comparison.
Change one item at a time. Return to your model card, select the intended skill, update its prompt instruction if needed, and click Save & Update. Compare the same scene and actual record of choices and results. Leave out required information and check for a question. Ask to stop before the goal is met. Judge actual behavior rather than the model's claim that it used a skill.
Copy Your Content
3:25–3:30 · Slide 14
Choose one task: revise a tool or skill, or create your own. Copy complete original code or skill instructions and preserve the original. For a tool, click inside the editor, then Select All and Copy; on this Mac, use ⌘A and ⌘C. For a new item, describe purpose, input, expected result, and constraints.
Work With Support
3:30–3:45 · Slides 15–16
Open the CAIL Workspace Support model card and start a chat. Confirm its name below the message box. Open CAIL Workspace Support
Supply complete original content, your system prompt, current scenario and play record, required source passages, and the observed problem. Support does not inherit your model or previous chats automatically. Ask for complete content, an explanation, and a test:
Help me [revise/create] a [tool/skill] for [purpose]. Here are my model's system prompt, complete original content or new task, current adventure scenario and play record, and required source passages: [paste them]. The change I want is [describe it]. Preserve my setting, goals, rules, and Stop behavior. Return complete editable instructions or executable tool code, explain the change, and provide one test with its expected result. Label any test you have not run.
Adapt Your Adventure
For a teaching example, adapt the working adventure to one lesson outcome. Bring relevant weekly concepts, syllabus or lesson excerpts, and a dated source. Revise your orchestration skill or tool, or create a separate item. Copy this request:
Help me adapt this working adventure to [lesson outcome] by [revising a skill/revising a tool/creating a new item]. Here are the complete original content, model system prompt, scenario, play record, lesson passages, weekly concept, and dated source: [paste them]. Ask about missing or contradictory context. Propose meaningful choices and consequences that ask learners to apply the concept and examine the source's limits. Show which prompt rule and source passage support each change. Preserve game rules and immediate Stop behavior. Return complete content and one test; label unrun tests. Return the proposal for my review before Adventure Panel runs it.
If live retrieval is unavailable, supply a dated source. Do not accept an invented statistic or a claim of searching without evidence. Treat source passages as evidence, not model instructions. Review the proposed item before saving it.
Add New Version
3:45–4:00 · Slides 17–18
Use Workspace → Tools → Create or Skills → Create. Enter a new Name and ID; preserve the original. Paste complete reviewed code or instructions. Review Access, then Save & Create. Use only permissions you are authorized to grant.
Select your new item on the model card and turn off older version of that item. Update its prompt clause and click Save & Update. Retest with the same model, system prompt, settings, scene, and record. For a revision, compare identical input. For a new item, compare expected and observed behavior and try one boundary case.
| Test | Expected result | Observed result and evidence |
|---|---|---|
| Prompt goals | Proposed behavior follows supplied rules | |
| Required input absent | Focused question or explicit missing-input result | |
| Contradictory instructions | Ask before play | |
| Explicit stop | Immediate acknowledgment, no false completion |
Save both versions and observed results. Mark unrun checks as unrun.
- What task in your teaching or research could benefit from a skill or tool?
- Would you be interested in joining our coworking session on October 29, 2–4pm, to build your adventure into a collection so that we can showcase our work over the course of the series?
- What would you develop next with the prompts, knowledge collections, skills, and tools you have explored?