Explorer Pipeline

Explorer Pipeline

Topic in, interactive 3D explorer out. The three skills in this bundle each do one job: the interface brief decides what the learner sees and does, the model prompt pack produces the parts and their sourced labels, and the viewer builder turns both into a page. This skill runs them in order for one topic, carries each step's output into the next, and keeps a build checklist so nothing ships unsourced or untested.

Required Inputs

Ask for these once, at the start, and reuse them in every stage:

• Topic and the object or system to explore
• Audience and learning goal
• Devices the explorer must work on
• Trusted sources the user has, or the curriculum or standard to follow
• The image-to-3D tool the user will use, or "not chosen yet"
• Hosting for the finished page

Output Structure

1. Run plan

A table of the three stages with their inputs, outputs and the person responsible for any manual step (generating images and meshes happens outside the chat): | Stage | Skill | Input | Output | Manual step |

2. Stage outputs

Run each stage in full, in order, with a short hand-off note between them:

1. explorer-interface-brief: the design brief and the image-model prompt for the mock. Hand-off: the implied parts list, palette and layout values.
2. model-prompt-pack: the prompt cards, the sourced parts list and the illustrative statement. Hand-off: mesh names, labels, descriptions and sources in the exact shape CONFIG.parts needs.
3. explorer-viewer-builder: the single-file viewer filled in from both hand-offs, with the demo assembly until the user's GLB exists.

Pause after stage 2 if the user must generate meshes first, and say exactly what to bring back (the GLB file and the mesh names it contains).

3. Build checklist

A checklist to keep until launch, grouped by stage:

markdownCopy code
#### Design
- [ ] Learning outcomes agreed with [teacher, client or reviewer]
- [ ] Interface mock generated and reviewed
#### Assets
- [ ] One reference image per part, checked against its source figure
- [ ] One mesh per part, named exactly as in the parts list
- [ ] GLB assembled, scaled, under 15 MB
#### Content
- [ ] Every label has a named trusted source
- [ ] Descriptions reviewed by a subject expert
- [ ] Illustrative statement visible on the page
#### Build and test
- [ ] Mesh-name check passes (no missing or extra names)
- [ ] Tested on desktop, a phone at 390 px width, and in dark mode
- [ ] Keyboard path to every part works
- [ ] No console errors
#### Publish
- [ ] Hosted at [URL]; link checked from a phone on mobile data

4. Status summary

At the end of each session: what is done, what is blocked (and on whom), and the next action.

Quality Checks

• All three stages are run in order, each with its full output
• Each hand-off passes the exact values the next stage needs
• The parts list from stage 2 is the one used in stage 3's CONFIG
• The pipeline pauses for manual mesh generation and says what to bring back
• The build checklist covers design, assets, content, testing and publishing
• No label reaches the viewer without a named source

Anti-Patterns

• Starting with the 3D model. Without the brief, the parts list and the page disagree.
• Losing the sources between stages. Sources travel with labels into the viewer, every time.
• Declaring it done before the mesh-name check. Mismatched names mean unclickable parts.
• Skipping the subject-expert review. Sourced labels can still be misread; a reviewer catches it.

Example Trigger Phrases

• "Make a 3D explorer of the human heart for Year 10 students, start to finish."
• "Topic in, 3D explorer out: jet engines for an aviation museum."
• "Run the whole 3D explorer pipeline for a plant cell."
• "I want an interactive 3D page about volcanoes. Take me through it."