## Feedback: Subagent Configuration & LaTeX Rendering Support

## Feedback: Subagent Configuration & LaTeX Rendering Support

### Feature Request 1: Subagent Configuration

**Current limitation:** The subagent system (explore, task, general-purpose, etc.) operates with fixed default behaviors and model selections. There is no way for users to configure subagent parameters such as:

- Default model selection per agent type (e.g., always use a specific model for `explore` agents)
- Custom system prompts or behavioral overrides
- Default context tier (`default` vs `long_context`) per agent type
- Reasoning effort defaults per agent type
- Timeout and max-turn configurations

**Suggested improvement:** Introduce a user-level or project-level configuration file (e.g., `.github/copilot-agents.json` or a section in `copilot-instructions.md`) that allows developers to tune subagent behavior. Example:

```json
{
"subagents": {
"explore": {
"model": "gpt-4o",
"context_tier": "long_context",
"reasoning_effort": "high"
},
"code-review": {
"model": "o3",
"reasoning_effort": "xhigh"
}
}
}
```

**Use case:** Different tasks benefit from different model capabilities. Code review benefits from high reasoning effort, while exploration can use a faster, cheaper model. Letting users configure this per agent type would significantly improve workflow efficiency.

---

### Feature Request 2: LaTeX Formula Rendering

**Current limitation:** When an agent produces mathematical content using LaTeX syntax (e.g., `$$\sum_{i=1}^{n} x_i$$` or inline `$E = mc^2$`), it is displayed as raw text in the chat interface.

**Suggested improvement:** Render LaTeX expressions in the chat UI:

- Block math: `$$...$$` → rendered as a centered block equation
- Inline math: `$...$` → rendered inline with surrounding text
- Use a well-tested library like MathJax or KaTeX for rendering

**Use case:** This is critical for:

- Technical discussions involving algorithms, statistics, or physics
- Academic and research-oriented coding tasks
- Any workflow where mathematical notation is part of the reasoning (e.g., complexity analysis, ML model definitions, signal processing)

---

### Summary

| Feature | Priority | Impact |
|---|---|---|
| Subagent configuration | High | Workflow efficiency, cost optimization, task-specific tuning |
| LaTeX rendering | Medium | Readability of mathematical content, better technical communication |

---
| Field | Value |
| --- | --- |
| App version | 1.1.10 |
| OS | Windows 10.0.26200 |
| Theme | GitHub |
| Path | /chat |
| Tenure | Week 10 |

The issue identifies no repository files, tests, or entry points. Start by locating the subagent configuration flow and chat rendering path, then clarify whether both requests belong in one change. Done would require agreed configuration semantics and tested rendering for both inline and block LaTeX.

由索引模型根据 Issue 内容生成。