Context-aware LLM prompt builder for node and edge panels

[paolino]

Summary

Each node and edge detail panel gets a text input + "Copy prompt" button. The user writes what they want (free text), and the tool assembles a complete prompt with all the context an LLM needs to do it.

UX

─────────────────────────────────
🔧 Ask an LLM about this node
┌───────────────────────────────┐
│ Update to reflect v0.15 ...   │
└───────────────────────────────┘
[Copy prompt]
─────────────────────────────────

The user provides intent. The tool provides context and process.

What "Copy prompt" assembles

1. User's text — verbatim, as the task
2. Current node/edge JSON — the data being discussed, plus connected edges
3. JSON schemas — links to the schema files so the LLM knows the exact format:
   • schema/graph.schema.json (nodes and edges)
   • schema/config.schema.json
   • schema/tutorial.schema.json and schema/tutorial-index.schema.json (if touching tutorials)
4. Source repo URL — from config.sourceUrl, so the LLM knows where to file a PR
5. File paths — data/graph.json, data/config.json, data/tutorials/ structure
6. PR instructions — how to fork, branch, edit the JSON, and submit

Key principle

The button is not "refresh this node." It's a context assembler — the user says what they want, the tool wraps it with everything needed for an LLM to execute and propose changes via PR.

Schema references

The schemas already exist in the repo:

• schema/graph.schema.json
• schema/config.schema.json
• schema/tutorial.schema.json
• schema/tutorial-index.schema.json
• schema/manifest.schema.json

The generated prompt should link to these via the raw GitHub URL derived from config.sourceUrl (e.g. https://raw.githubusercontent.com/{owner}/{repo}/main/schema/graph.schema.json). If sourceUrl points to a graph-browser consumer repo (not graph-browser itself), the schema URLs should point to graph-browser's repo instead.

Works for any dataset

Nothing is hardcoded to a specific graph. The prompt is built from config.sourceUrl + the selected node/edge data + the schema URLs.