Copilot instructions

1<!-- llm-router-rules-version: 1 -->
2# LLM Router — VS Code / GitHub Copilot Routing Rules
3 
4> These rules apply when using llm-router MCP tools inside VS Code with GitHub Copilot.
5> VS Code loads MCP servers from .vscode/mcp.json (workspace) or the user mcp.json.
6> Use `llm_auto` for routing + cross-session savings tracking.
7 
8---
9 
10## How to Route
11 
12VS Code loads MCP servers on startup. Once llm-router is registered, call tools by name:
13 
14```
15Rule: for research/generate/code/analyze tasks → call llm_auto(prompt=<the task>)
16```
17 
18Add to `.github/copilot-instructions.md` for workspace-level routing guidance:
19 
20```
21When a task requires live web search, call the llm_research MCP tool.
22For code generation or complex analysis, call llm_code or llm_analyze.
23For auto-routing with savings tracking, call llm_auto.
24```
25 
26---
27 
28## Task Type → Tool Mapping
29 
30| Task | Tool | Notes |
31|---|---|---|
32| Simple factual question | `llm_query` | Gemini Flash / Groq |
33| Research / current events | `llm_research` | Perplexity (web-grounded) |
34| Writing / content | `llm_generate` | Gemini Flash / Haiku |
35| Deep analysis | `llm_analyze` | GPT-4o / Gemini Pro |
36| Code generation | `llm_code` | Ollama → Codex → o3 |
37| Unknown / auto-detect | `llm_auto` | Classifies + routes |
38 
39---
40 
41## Savings Visibility
42 
43Run `llm_savings` in Copilot Chat to see cross-session totals.
44 
45---
46 
47## Token-Efficient Responses
48 
49Skip preamble. Lead with result. Fragments fine when meaning is clear.
50No trailing summaries. ≥3 items → bullets. Never restate the user's request.
51